diff --git a/backends/arm/TARGETS b/backends/arm/TARGETS
index 220db373710..39910f01503 100644
--- a/backends/arm/TARGETS
+++ b/backends/arm/TARGETS
@@ -8,7 +8,7 @@ python_library(
     typing = True,
     deps = [
         ":arm_backend",
-        "//executorch/backends/arm/passes:passes",
+        "//executorch/backends/arm/_passes:passes",
         "//executorch/exir:lib",
     ],
 )
@@ -27,7 +27,7 @@ python_library(
         ":arm_vela",
         "//executorch/backends/arm/operators:lib",
         "//executorch/backends/arm/operators:node_visitor",
-        "//executorch/backends/arm/passes:passes",
+        "//executorch/backends/arm/_passes:passes",
     ],
 )
 
diff --git a/backends/arm/passes/TARGETS b/backends/arm/_passes/TARGETS
similarity index 100%
rename from backends/arm/passes/TARGETS
rename to backends/arm/_passes/TARGETS
diff --git a/backends/arm/passes/annotate_channels_last_dim_order_pass.py b/backends/arm/_passes/annotate_channels_last_dim_order_pass.py
similarity index 100%
rename from backends/arm/passes/annotate_channels_last_dim_order_pass.py
rename to backends/arm/_passes/annotate_channels_last_dim_order_pass.py
diff --git a/backends/arm/passes/arm_pass_manager.py b/backends/arm/_passes/arm_pass_manager.py
similarity index 87%
rename from backends/arm/passes/arm_pass_manager.py
rename to backends/arm/_passes/arm_pass_manager.py
index 75ef551171e..620109ef604 100644
--- a/backends/arm/passes/arm_pass_manager.py
+++ b/backends/arm/_passes/arm_pass_manager.py
@@ -8,13 +8,13 @@
 # pyre-unsafe
 
 import torch
-from executorch.backends.arm.passes.annotate_channels_last_dim_order_pass import (
+from executorch.backends.arm._passes.annotate_channels_last_dim_order_pass import (
     AnnotateChannelsLastDimOrder,
 )
-from executorch.backends.arm.passes.convert_expand_copy_to_repeat import (
+from executorch.backends.arm._passes.convert_expand_copy_to_repeat import (
     ConvertExpandCopyToRepeatPass,
 )
-from executorch.backends.arm.passes.convert_split_to_slice import (
+from executorch.backends.arm._passes.convert_split_to_slice import (
     ConvertSplitToSlicePass,
 )
 from executorch.backends.arm.passes.meandim_to_averagepool_pass import (
diff --git a/backends/arm/_passes/arm_pass_utils.py b/backends/arm/_passes/arm_pass_utils.py
new file mode 100644
index 00000000000..34704d2cedf
--- /dev/null
+++ b/backends/arm/_passes/arm_pass_utils.py
@@ -0,0 +1,66 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import Optional
+
+import torch
+
+from executorch.exir.dialects._ops import ops as exir_ops
+from torch._ops import OpOverload
+
+
+def create_node(
+    graph: torch.fx.Graph,
+    op_target: OpOverload,
+    args: tuple = (),
+    kwargs: Optional[dict] = None,
+    quantize: bool = False,
+    q_params: Optional[tuple] = None,
+):
+    """
+    Adds a node to 'graph'. graph.inserting_before/after() should be used before the call to decide where to insert the node.
+    If quantize is true and q_params is not None, a q dq pair is inserted after the newly created node.
+    """
+
+    node = graph.create_node(
+        "call_function",
+        op_target,
+        args=args,
+        kwargs=kwargs or {},
+    )
+    if quantize and q_params:
+        return insert_q_dq_pair(graph, node, q_params)
+    return node
+
+
+def insert_q_dq_pair(
+    graph: torch.fx.Graph,
+    anchor: torch.fx.Node,
+    q_params: tuple,
+):
+    """
+    Inserts a q dq node pair after the node 'anchor'.
+    """
+
+    with graph.inserting_after(anchor):
+        q = create_node(
+            graph=graph,
+            op_target=exir_ops.edge.quantized_decomposed.quantize_per_tensor.default,
+            args=(),  # We add the argument last
+        )
+        q.meta = anchor.meta
+    with graph.inserting_after(q):
+        dq = create_node(
+            graph=graph,
+            op_target=exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default,
+            args=(q,) + q_params,
+        )
+        dq.meta = q.meta
+    anchor.replace_all_uses_with(dq)
+    # We add this last so the replace all uses above does not replace the quantized
+    # node's first use
+    q.args = (anchor,) + q_params
+    return dq
diff --git a/backends/arm/_passes/cast_int64_pass.py b/backends/arm/_passes/cast_int64_pass.py
new file mode 100644
index 00000000000..6bdbca62879
--- /dev/null
+++ b/backends/arm/_passes/cast_int64_pass.py
@@ -0,0 +1,35 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from executorch.exir.pass_base import ExportPass, PassResult
+
+
+class CastInt64ToInt32Pass(ExportPass):
+    def __init__(self, exported_program: torch.export.ExportedProgram):
+        super(CastInt64ToInt32Pass, self).__init__()
+        self.exported_program = exported_program
+
+    def _to_int32(self, graph_module: torch.fx.GraphModule):
+        for node in graph_module.graph.nodes:
+            fake_tensor = node.meta["val"]
+            if isinstance(fake_tensor, torch._subclasses.fake_tensor.FakeTensor):
+                if node.meta["val"].dtype == torch.int64:
+                    node.meta["val"] = node.meta["val"].to(torch.int32)
+                    buffer_name = (
+                        self.exported_program.graph_signature.inputs_to_buffers[
+                            node.name
+                        ]
+                    )
+                    new_tensor = self.exported_program.state_dict[buffer_name].to(
+                        torch.int32
+                    )
+                    self.exported_program.state_dict[buffer_name] = new_tensor
+
+    def call(self, graph_module: torch.fx.GraphModule):
+        self._to_int32(graph_module)
+        graph_module.recompile()
+        graph_module = super().call(graph_module).graph_module
+        return PassResult(graph_module, True)
diff --git a/backends/arm/passes/convert_expand_copy_to_repeat.py b/backends/arm/_passes/convert_expand_copy_to_repeat.py
similarity index 100%
rename from backends/arm/passes/convert_expand_copy_to_repeat.py
rename to backends/arm/_passes/convert_expand_copy_to_repeat.py
diff --git a/backends/arm/passes/convert_split_to_slice.py b/backends/arm/_passes/convert_split_to_slice.py
similarity index 100%
rename from backends/arm/passes/convert_split_to_slice.py
rename to backends/arm/_passes/convert_split_to_slice.py
diff --git a/backends/arm/_passes/decompose_div_pass.py b/backends/arm/_passes/decompose_div_pass.py
new file mode 100644
index 00000000000..13ee8d8dff7
--- /dev/null
+++ b/backends/arm/_passes/decompose_div_pass.py
@@ -0,0 +1,45 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from executorch.exir.dialects._ops import ops as exir_ops
+from executorch.exir.pass_base import ExportPass
+
+
+def get_div_decomposition(op) -> tuple:
+    """
+    Returns the the (reciprocal_op, mul_op), where the ops depends on if
+    the div op is in exir_ops torch.ops.aten.
+    """
+    if op == exir_ops.edge.aten.div.Tensor:
+        return (exir_ops.edge.aten.reciprocal.default, exir_ops.edge.aten.mul.Tensor)
+    if op == torch.ops.aten.div.Tensor:
+        return (torch.ops.aten.reciprocal.default, torch.ops.aten.mul.Tensor)
+    raise RuntimeError(f"Can't get div decomposition for op {op}")
+
+
+class DecomposeDivPass(ExportPass):
+    """
+    This pass decomposes div into a mul and a reciprocal node.
+
+    Example:
+        y = div(a,b)
+    Becomes:
+        x = reciprocal(b)
+        y = mul(a,x)
+    """
+
+    def call_operator(self, op, args, kwargs, meta):
+        if op not in (exir_ops.edge.aten.div.Tensor, torch.ops.aten.div.Tensor):
+            return super().call_operator(op, args, kwargs, meta)
+
+        reciprocal_op, mul_op = get_div_decomposition(op)
+
+        numerator = args[0]
+        denominator = args[1]
+        reciprocal = super().call_operator(reciprocal_op, (denominator,), {}, meta)
+
+        return super().call_operator(mul_op, (numerator, reciprocal), {}, meta)
diff --git a/backends/arm/passes/meandim_to_averagepool_pass.py b/backends/arm/_passes/meandim_to_averagepool_pass.py
similarity index 100%
rename from backends/arm/passes/meandim_to_averagepool_pass.py
rename to backends/arm/_passes/meandim_to_averagepool_pass.py
diff --git a/backends/arm/passes/remove_clone_pass.py b/backends/arm/_passes/remove_clone_pass.py
similarity index 100%
rename from backends/arm/passes/remove_clone_pass.py
rename to backends/arm/_passes/remove_clone_pass.py
diff --git a/backends/arm/_passes/scalars_to_attribute_pass.py b/backends/arm/_passes/scalars_to_attribute_pass.py
new file mode 100644
index 00000000000..e9e547b9c96
--- /dev/null
+++ b/backends/arm/_passes/scalars_to_attribute_pass.py
@@ -0,0 +1,69 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import cast, Union
+
+import torch
+from executorch.backends.arm.tosa_mapping import extract_tensor_meta
+
+from executorch.exir.pass_base import ExportPass, PassResult
+from torch.ao.quantization.fx.utils import get_new_attr_name_with_prefix
+from torch.fx import GraphModule, Node
+
+
+class ScalarsToAttributePass(ExportPass):
+    """
+    For ops in 'targeted_ops', convert inputs that are scalar values
+    to attribute Nodes that output the same value.
+    """
+
+    targeted_ops = [
+        torch.ops.aten.add.Tensor,
+        torch.ops.aten.sub.Tensor,
+        torch.ops.aten.sub_.Tensor,
+        torch.ops.aten.mul.Tensor,
+        torch.ops.aten.div.Tensor,
+    ]
+
+    def call(self, graph_module: GraphModule) -> PassResult:
+        for n in graph_module.graph.nodes:
+            n = cast(Node, n)
+            if n.op != "call_function" or n.target not in self.targeted_ops:
+                continue
+
+            biggest_rank = 1
+            for arg in n.args:
+                if isinstance(arg, Node):
+                    _, shape, _ = extract_tensor_meta(arg.meta)
+                    biggest_rank = max(biggest_rank, len(shape))
+
+            new_args = []
+            for arg in n.args:
+                if isinstance(arg, Node):
+                    new_args.append(arg)
+                    continue
+
+                prefix = "_tensor_constant_"
+                get_new_attr_name = get_new_attr_name_with_prefix(prefix)
+                tensor_constant_name = get_new_attr_name(graph_module)
+                float_tensor = torch.tensor(
+                    float(cast(Union[int, float], arg))
+                ).reshape((1,) * biggest_rank)
+                graph_module.register_buffer(tensor_constant_name, float_tensor)
+                fake_mode = n.meta["val"].fake_mode
+
+                with graph_module.graph.inserting_before(n):
+                    get_attr_node = graph_module.graph.create_node(
+                        "get_attr", tensor_constant_name, (), {}
+                    )
+                    get_attr_node.meta["val"] = fake_mode.from_tensor(
+                        float_tensor, static_shapes=True
+                    )
+                    new_args.append(get_attr_node)
+            n.args = tuple(new_args)
+
+        graph_module.recompile()
+        return PassResult(graph_module, True)
diff --git a/backends/arm/passes/size_adjust_conv2d_pass.py b/backends/arm/_passes/size_adjust_conv2d_pass.py
similarity index 100%
rename from backends/arm/passes/size_adjust_conv2d_pass.py
rename to backends/arm/_passes/size_adjust_conv2d_pass.py
diff --git a/backends/arm/passes/tag_io_quant_pass.py b/backends/arm/_passes/tag_io_quant_pass.py
similarity index 100%
rename from backends/arm/passes/tag_io_quant_pass.py
rename to backends/arm/_passes/tag_io_quant_pass.py
diff --git a/backends/arm/arm_backend.py b/backends/arm/arm_backend.py
index a5f47c222fe..c3e2e84c6ac 100644
--- a/backends/arm/arm_backend.py
+++ b/backends/arm/arm_backend.py
@@ -20,7 +20,9 @@
 from executorch.backends.arm.operators.node_visitor import get_node_visitors
 from executorch.backends.arm.operators.op_output import process_output
 from executorch.backends.arm.operators.op_placeholder import process_placeholder
-from executorch.backends.arm.passes.arm_pass_manager import ArmPassManager
+from executorch.backends.arm._passes.arm_pass_manager import (
+    ArmPassManager,
+)  # usort: skip
 from executorch.backends.arm.tosa_utils import (
     dbg_fail,
     dbg_tosa_dump,
diff --git a/backends/arm/arm_partitioner.py b/backends/arm/arm_partitioner.py
index 6b57c3d9658..7793060f6f9 100644
--- a/backends/arm/arm_partitioner.py
+++ b/backends/arm/arm_partitioner.py
@@ -11,8 +11,8 @@
 from typing import final, List
 
 import torch
-from executorch.backends.arm.arm_backend import ArmBackend
-from executorch.backends.arm.passes.tag_io_quant_pass import TagIOQuantPass
+from executorch.backends.arm.arm_backend import ArmBackend  # usort: skip
+from executorch.backends.arm._passes.tag_io_quant_pass import TagIOQuantPass
 from executorch.exir.backend.compile_spec_schema import CompileSpec
 from executorch.exir.backend.partitioner import (
     DelegationSpec,
diff --git a/backends/arm/test/passes/test_meandim_to_averagepool2d.py b/backends/arm/test/passes/test_meandim_to_averagepool2d.py
index 1cd63e6e52e..c8fa0f4b7ab 100644
--- a/backends/arm/test/passes/test_meandim_to_averagepool2d.py
+++ b/backends/arm/test/passes/test_meandim_to_averagepool2d.py
@@ -7,7 +7,7 @@
 import unittest
 
 import torch
-from executorch.backends.arm.passes.meandim_to_averagepool_pass import (
+from executorch.backends.arm._passes.meandim_to_averagepool_pass import (
     ConvertMeanDimToAveragePool,
 )
 
diff --git a/backends/qualcomm/README.md b/backends/qualcomm/README.md
index be7cd427d6e..d426aff16b1 100644
--- a/backends/qualcomm/README.md
+++ b/backends/qualcomm/README.md
@@ -46,7 +46,7 @@ backends/qualcomm
 |   └── python # Python interface for using QNN libraries.
 ├── builders # Codes for lowering each operators (AoT Part).
 ├── partition # QNN Partitioner (AoT Part).
-├── passes # Various passes helping lower models to QNN backend (AoT Part).
+├── _passes # Various private passes helping lower models to QNN backend (AoT Part).
 ├── python # Places to put pybind artifacts for accessing QNN APIs, structures, etc (AoT Part).
 ├── quantizer # QNN Quantizer
 ├── runtime # Here is QNN runtime responsbile for compiling a model on x64.
diff --git a/backends/qualcomm/_passes/TARGETS b/backends/qualcomm/_passes/TARGETS
new file mode 100644
index 00000000000..a824ca9f6e5
--- /dev/null
+++ b/backends/qualcomm/_passes/TARGETS
@@ -0,0 +1,18 @@
+load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
+
+oncall("executorch")
+
+runtime.python_library(
+    name = "passes",
+    srcs = glob([
+        "*.py",
+    ]),
+    visibility = [
+        "@EXECUTORCH_CLIENTS",
+    ],
+    deps = [
+        "//executorch/backends/transforms:addmm_mm_to_linear",
+        "//executorch/exir/backend:backend_details",
+        "//executorch/exir/backend:compile_spec_schema",
+    ],
+)
diff --git a/backends/qualcomm/passes/annotate_and_quant_scalar.py b/backends/qualcomm/_passes/annotate_and_quant_scalar.py
similarity index 100%
rename from backends/qualcomm/passes/annotate_and_quant_scalar.py
rename to backends/qualcomm/_passes/annotate_and_quant_scalar.py
diff --git a/backends/qualcomm/passes/annotate_decomposed.py b/backends/qualcomm/_passes/annotate_decomposed.py
similarity index 100%
rename from backends/qualcomm/passes/annotate_decomposed.py
rename to backends/qualcomm/_passes/annotate_decomposed.py
diff --git a/backends/qualcomm/passes/annotate_quant_attrs.py b/backends/qualcomm/_passes/annotate_quant_attrs.py
similarity index 100%
rename from backends/qualcomm/passes/annotate_quant_attrs.py
rename to backends/qualcomm/_passes/annotate_quant_attrs.py
diff --git a/backends/qualcomm/passes/build_quant_io.py b/backends/qualcomm/_passes/build_quant_io.py
similarity index 100%
rename from backends/qualcomm/passes/build_quant_io.py
rename to backends/qualcomm/_passes/build_quant_io.py
diff --git a/backends/qualcomm/passes/convert_binary_op_with_scalar.py b/backends/qualcomm/_passes/convert_binary_op_with_scalar.py
similarity index 100%
rename from backends/qualcomm/passes/convert_binary_op_with_scalar.py
rename to backends/qualcomm/_passes/convert_binary_op_with_scalar.py
diff --git a/backends/qualcomm/passes/convert_bmm_to_matmul.py b/backends/qualcomm/_passes/convert_bmm_to_matmul.py
similarity index 100%
rename from backends/qualcomm/passes/convert_bmm_to_matmul.py
rename to backends/qualcomm/_passes/convert_bmm_to_matmul.py
diff --git a/backends/qualcomm/passes/convert_interpolate_with_upsample2d.py b/backends/qualcomm/_passes/convert_interpolate_with_upsample2d.py
similarity index 100%
rename from backends/qualcomm/passes/convert_interpolate_with_upsample2d.py
rename to backends/qualcomm/_passes/convert_interpolate_with_upsample2d.py
diff --git a/backends/qualcomm/passes/convert_prelu.py b/backends/qualcomm/_passes/convert_prelu.py
similarity index 100%
rename from backends/qualcomm/passes/convert_prelu.py
rename to backends/qualcomm/_passes/convert_prelu.py
diff --git a/backends/qualcomm/passes/convert_to_linear.py b/backends/qualcomm/_passes/convert_to_linear.py
similarity index 100%
rename from backends/qualcomm/passes/convert_to_linear.py
rename to backends/qualcomm/_passes/convert_to_linear.py
diff --git a/backends/qualcomm/passes/decompose_silu.py b/backends/qualcomm/_passes/decompose_silu.py
similarity index 100%
rename from backends/qualcomm/passes/decompose_silu.py
rename to backends/qualcomm/_passes/decompose_silu.py
diff --git a/backends/qualcomm/passes/fold_qdq.py b/backends/qualcomm/_passes/fold_qdq.py
similarity index 100%
rename from backends/qualcomm/passes/fold_qdq.py
rename to backends/qualcomm/_passes/fold_qdq.py
diff --git a/backends/qualcomm/passes/fuse_consecutive_transpose.py b/backends/qualcomm/_passes/fuse_consecutive_transpose.py
similarity index 100%
rename from backends/qualcomm/passes/fuse_consecutive_transpose.py
rename to backends/qualcomm/_passes/fuse_consecutive_transpose.py
diff --git a/backends/qualcomm/passes/i64_to_i32.py b/backends/qualcomm/_passes/i64_to_i32.py
similarity index 100%
rename from backends/qualcomm/passes/i64_to_i32.py
rename to backends/qualcomm/_passes/i64_to_i32.py
diff --git a/backends/qualcomm/passes/insert_io_qdq.py b/backends/qualcomm/_passes/insert_io_qdq.py
similarity index 100%
rename from backends/qualcomm/passes/insert_io_qdq.py
rename to backends/qualcomm/_passes/insert_io_qdq.py
diff --git a/backends/qualcomm/passes/insert_requantize.py b/backends/qualcomm/_passes/insert_requantize.py
similarity index 100%
rename from backends/qualcomm/passes/insert_requantize.py
rename to backends/qualcomm/_passes/insert_requantize.py
diff --git a/backends/qualcomm/passes/layout_transform.py b/backends/qualcomm/_passes/layout_transform.py
similarity index 100%
rename from backends/qualcomm/passes/layout_transform.py
rename to backends/qualcomm/_passes/layout_transform.py
diff --git a/backends/qualcomm/passes/recompose_pixel_unshuffle.py b/backends/qualcomm/_passes/recompose_pixel_unshuffle.py
similarity index 100%
rename from backends/qualcomm/passes/recompose_pixel_unshuffle.py
rename to backends/qualcomm/_passes/recompose_pixel_unshuffle.py
diff --git a/backends/qualcomm/passes/recompose_rms_norm.py b/backends/qualcomm/_passes/recompose_rms_norm.py
similarity index 100%
rename from backends/qualcomm/passes/recompose_rms_norm.py
rename to backends/qualcomm/_passes/recompose_rms_norm.py
diff --git a/backends/qualcomm/passes/reduce_dynamic_range.py b/backends/qualcomm/_passes/reduce_dynamic_range.py
similarity index 100%
rename from backends/qualcomm/passes/reduce_dynamic_range.py
rename to backends/qualcomm/_passes/reduce_dynamic_range.py
diff --git a/backends/qualcomm/passes/remove_redundancy.py b/backends/qualcomm/_passes/remove_redundancy.py
similarity index 100%
rename from backends/qualcomm/passes/remove_redundancy.py
rename to backends/qualcomm/_passes/remove_redundancy.py
diff --git a/backends/qualcomm/passes/replace_index_put_input.py b/backends/qualcomm/_passes/replace_index_put_input.py
similarity index 100%
rename from backends/qualcomm/passes/replace_index_put_input.py
rename to backends/qualcomm/_passes/replace_index_put_input.py
diff --git a/backends/qualcomm/passes/replace_inf_buffer.py b/backends/qualcomm/_passes/replace_inf_buffer.py
similarity index 100%
rename from backends/qualcomm/passes/replace_inf_buffer.py
rename to backends/qualcomm/_passes/replace_inf_buffer.py
diff --git a/backends/qualcomm/passes/utils.py b/backends/qualcomm/_passes/utils.py
similarity index 100%
rename from backends/qualcomm/passes/utils.py
rename to backends/qualcomm/_passes/utils.py
diff --git a/backends/qualcomm/qnn_preprocess.py b/backends/qualcomm/qnn_preprocess.py
index 6360b35280f..417934acbd4 100644
--- a/backends/qualcomm/qnn_preprocess.py
+++ b/backends/qualcomm/qnn_preprocess.py
@@ -11,15 +11,15 @@
 import executorch.backends.qualcomm.python.PyQnnManagerAdaptor as PyQnnManager
 
 import torch  # noqa: F401
-from executorch.backends.qualcomm.builders.node_visitor import get_node_visitors
-from executorch.backends.qualcomm.builders.qnn_constants import OpContextLoader
-from executorch.backends.qualcomm.passes.convert_to_linear import ConvertToLinear
-from executorch.backends.qualcomm.passes.fuse_consecutive_transpose import (
+from executorch.backends.qualcomm._passes.convert_to_linear import ConvertToLinear
+from executorch.backends.qualcomm._passes.fuse_consecutive_transpose import (
     FuseConsecutiveTranspose,
 )
-from executorch.backends.qualcomm.passes.insert_io_qdq import InsertIOQDQ
-from executorch.backends.qualcomm.passes.insert_requantize import InsertRequantize
-from executorch.backends.qualcomm.passes.layout_transform import LayoutTransform
+from executorch.backends.qualcomm._passes.insert_io_qdq import InsertIOQDQ
+from executorch.backends.qualcomm._passes.insert_requantize import InsertRequantize
+from executorch.backends.qualcomm._passes.layout_transform import LayoutTransform
+from executorch.backends.qualcomm.builders.node_visitor import get_node_visitors
+from executorch.backends.qualcomm.builders.qnn_constants import OpContextLoader
 from executorch.backends.qualcomm.utils.utils import generate_qnn_executorch_option
 from executorch.exir.backend.backend_details import (
     BackendDetails,
diff --git a/backends/qualcomm/quantizer/quantizer.py b/backends/qualcomm/quantizer/quantizer.py
index e27edf939c8..b479d79ab89 100644
--- a/backends/qualcomm/quantizer/quantizer.py
+++ b/backends/qualcomm/quantizer/quantizer.py
@@ -7,12 +7,12 @@
 from typing import Callable, Dict, Optional, Sequence, Set
 
 import torch
-from executorch.backends.qualcomm.passes.decompose_silu import DecomposeSilu
-from executorch.backends.qualcomm.passes.recompose_pixel_unshuffle import (
+from executorch.backends.qualcomm._passes.decompose_silu import DecomposeSilu
+from executorch.backends.qualcomm._passes.recompose_pixel_unshuffle import (
     RecomposePixelUnshuffle,
 )
-from executorch.backends.qualcomm.passes.reduce_dynamic_range import ReduceDynamicRange
-from executorch.backends.qualcomm.passes.replace_inf_buffer import ReplaceInfBuffer
+from executorch.backends.qualcomm._passes.reduce_dynamic_range import ReduceDynamicRange
+from executorch.backends.qualcomm._passes.replace_inf_buffer import ReplaceInfBuffer
 from executorch.backends.transforms.decompose_sdpa import (
     DecomposeScaledDotProductAttention,
 )
diff --git a/backends/qualcomm/utils/constants.py b/backends/qualcomm/utils/constants.py
index 9875c9f5afb..4787e89ca64 100644
--- a/backends/qualcomm/utils/constants.py
+++ b/backends/qualcomm/utils/constants.py
@@ -6,7 +6,7 @@
 
 # Qualcomm specific key
 
-# constants in backends/qualcomm/passes & backends/qualcomm/builders
+# constants in backends/qualcomm/_passes & backends/qualcomm/builders
 QCOM_AXIS = "axis"
 QCOM_AXIS_ORDER = "axis_order"
 QCOM_BITWIDTH = "bitwidth"
diff --git a/backends/qualcomm/utils/utils.py b/backends/qualcomm/utils/utils.py
index e1be24d0d64..4509843abac 100644
--- a/backends/qualcomm/utils/utils.py
+++ b/backends/qualcomm/utils/utils.py
@@ -14,37 +14,42 @@
 import executorch.exir as exir
 
 import torch
-
-from executorch.backends.qualcomm.builders.node_visitor import (
-    QNN_QUANT_TYPE_MAP,
-    QNN_TENSOR_TYPE_MAP,
-)
-from executorch.backends.qualcomm.builders.qnn_constants import OpContextLoader
-from executorch.backends.qualcomm.passes.annotate_and_quant_scalar import (
+from executorch.backends.qualcomm._passes.annotate_and_quant_scalar import (
     AnnotateAndQuantScalar,
 )
-from executorch.backends.qualcomm.passes.annotate_decomposed import AnnotateDecomposed
-from executorch.backends.qualcomm.passes.annotate_quant_attrs import AnnotateQuantAttrs
-from executorch.backends.qualcomm.passes.convert_binary_op_with_scalar import (
+from executorch.backends.qualcomm._passes.annotate_decomposed import AnnotateDecomposed
+from executorch.backends.qualcomm._passes.annotate_quant_attrs import AnnotateQuantAttrs
+from executorch.backends.qualcomm._passes.convert_binary_op_with_scalar import (
     ConvertBinaryOpsWithScalar,
 )
-from executorch.backends.qualcomm.passes.convert_bmm_to_matmul import ConvertBmmToMatmul
-from executorch.backends.qualcomm.passes.convert_interpolate_with_upsample2d import (
+from executorch.backends.qualcomm._passes.convert_bmm_to_matmul import (
+    ConvertBmmToMatmul,
+)
+from executorch.backends.qualcomm._passes.convert_interpolate_with_upsample2d import (
     ConvertInterpolateWithUpsample2D,
 )
-from executorch.backends.qualcomm.passes.convert_prelu import ConvertPReLU
-from executorch.backends.qualcomm.passes.convert_to_linear import ConvertToLinear
-from executorch.backends.qualcomm.passes.fold_qdq import FoldQDQ
-from executorch.backends.qualcomm.passes.i64_to_i32 import I64toI32
-from executorch.backends.qualcomm.passes.layout_transform import LayoutTransform
-from executorch.backends.qualcomm.passes.recompose_pixel_unshuffle import (
+from executorch.backends.qualcomm._passes.convert_prelu import ConvertPReLU
+from executorch.backends.qualcomm._passes.convert_to_linear import ConvertToLinear
+from executorch.backends.qualcomm._passes.fold_qdq import FoldQDQ
+from executorch.backends.qualcomm._passes.i64_to_i32 import I64toI32
+from executorch.backends.qualcomm._passes.layout_transform import LayoutTransform
+from executorch.backends.qualcomm._passes.recompose_pixel_unshuffle import (
     RecomposePixelUnshuffle,
 )
-from executorch.backends.qualcomm.passes.recompose_rms_norm import RecomposeRmsNorm
-from executorch.backends.qualcomm.passes.remove_redundancy import RemoveRedundancy
-from executorch.backends.qualcomm.passes.replace_index_put_input import (
+from executorch.backends.qualcomm._passes.recompose_rms_norm import RecomposeRmsNorm
+from executorch.backends.qualcomm._passes.remove_redundancy import RemoveRedundancy
+from executorch.backends.qualcomm._passes.replace_index_put_input import (
     ReplaceIndexPutInput,
 )
+
+from executorch.backends.qualcomm.builders.node_visitor import (
+    QNN_QUANT_TYPE_MAP,
+    QNN_TENSOR_TYPE_MAP,
+)
+from executorch.backends.qualcomm.builders.qnn_constants import OpContextLoader
+from executorch.backends.qualcomm.passes.expand_broadcast_tensor_shape import (
+    ExpandBroadcastTensorShape,
+)
 from executorch.backends.qualcomm.serialization.qnn_compile_spec_schema import (
     _soc_info_table,
     QcomChipset,
diff --git a/examples/qualcomm/oss_scripts/llama2/llama.py b/examples/qualcomm/oss_scripts/llama2/llama.py
index 9712b1c08fd..cd7ec24d3a7 100644
--- a/examples/qualcomm/oss_scripts/llama2/llama.py
+++ b/examples/qualcomm/oss_scripts/llama2/llama.py
@@ -12,9 +12,9 @@
 from multiprocessing.connection import Client
 
 import torch
+from executorch.backends.qualcomm._passes.build_quant_io import BuildQuantIo
 
 from executorch.backends.qualcomm.partition.qnn_partitioner import QnnPartitioner
-from executorch.backends.qualcomm.passes.build_quant_io import BuildQuantIo
 
 from executorch.backends.qualcomm.quantizer.quantizer import QuantDtype
 from executorch.backends.qualcomm.serialization.qnn_compile_spec_schema import (