diff --git a/backends/aoti/common_shims_slim.h b/backends/aoti/common_shims_slim.h
index b8948c0a311..dfcdecd2bc2 100644
--- a/backends/aoti/common_shims_slim.h
+++ b/backends/aoti/common_shims_slim.h
@@ -18,7 +18,7 @@
 // CUDA backend (SlimTensor) and other backends like MPS (ETensor).
 // The caller determines which path is used by defining CUDA_AVAILABLE.
 #ifdef CUDA_AVAILABLE
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
 #else
 #include <executorch/runtime/core/exec_aten/exec_aten.h>
 #endif
diff --git a/backends/aoti/slim/core/SlimTensor.h b/backends/aoti/slim/core/slim_tensor.h
similarity index 98%
rename from backends/aoti/slim/core/SlimTensor.h
rename to backends/aoti/slim/core/slim_tensor.h
index 226776d2bc8..b32186152e2 100644
--- a/backends/aoti/slim/core/SlimTensor.h
+++ b/backends/aoti/slim/core/slim_tensor.h
@@ -20,9 +20,9 @@
 #include <executorch/backends/aoti/slim/c10/core/Device.h>
 #include <executorch/backends/aoti/slim/c10/core/ScalarType.h>
 #include <executorch/backends/aoti/slim/c10/core/SizesAndStrides.h>
-#include <executorch/backends/aoti/slim/core/Storage.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
-#include <executorch/backends/aoti/slim/util/SizeUtil.h>
+#include <executorch/backends/aoti/slim/core/storage.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
+#include <executorch/backends/aoti/slim/util/size_util.h>
 #include <executorch/runtime/platform/assert.h>
 
 namespace executorch::backends::aoti::slim {
@@ -568,4 +568,4 @@ class SlimTensor {
 
 // Include view operations implementations (must be after SlimTensor class
 // definition)
-#include <executorch/backends/aoti/slim/core/SlimTensorView-incl.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor_view_incl.h>
diff --git a/backends/aoti/slim/core/SlimTensorView-incl.h b/backends/aoti/slim/core/slim_tensor_view_incl.h
similarity index 98%
rename from backends/aoti/slim/core/SlimTensorView-incl.h
rename to backends/aoti/slim/core/slim_tensor_view_incl.h
index a437f187299..82b605fb569 100644
--- a/backends/aoti/slim/core/SlimTensorView-incl.h
+++ b/backends/aoti/slim/core/slim_tensor_view_incl.h
@@ -9,7 +9,7 @@
 #pragma once
 
 #include <executorch/backends/aoti/slim/c10/core/WrapDimMinimal.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/core/Storage.h b/backends/aoti/slim/core/storage.h
similarity index 98%
rename from backends/aoti/slim/core/Storage.h
rename to backends/aoti/slim/core/storage.h
index 156556aa9e1..0ef1b204484 100644
--- a/backends/aoti/slim/core/Storage.h
+++ b/backends/aoti/slim/core/storage.h
@@ -17,9 +17,9 @@
 
 #include <executorch/backends/aoti/slim/c10/core/Device.h>
 #include <executorch/backends/aoti/slim/c10/core/ScalarType.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
-#include <executorch/backends/aoti/slim/util/SharedPtr.h>
-#include <executorch/backends/aoti/slim/util/SizeUtil.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
+#include <executorch/backends/aoti/slim/util/shared_ptr.h>
+#include <executorch/backends/aoti/slim/util/size_util.h>
 #include <executorch/runtime/platform/assert.h>
 #include <executorch/runtime/platform/log.h>
 
diff --git a/backends/aoti/slim/core/targets.bzl b/backends/aoti/slim/core/targets.bzl
index 408738edd35..735e8b52e1b 100644
--- a/backends/aoti/slim/core/targets.bzl
+++ b/backends/aoti/slim/core/targets.bzl
@@ -7,7 +7,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "storage",
         headers = [
-            "Storage.h",
+            "storage.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
@@ -25,8 +25,8 @@ def define_common_targets():
     runtime.cxx_library(
         name = "slimtensor",
         headers = [
-            "SlimTensor.h",
-            "SlimTensorView-incl.h",
+            "slim_tensor.h",
+            "slim_tensor_view_incl.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
diff --git a/backends/aoti/slim/core/test/test_as_strided.cpp b/backends/aoti/slim/core/test/test_as_strided.cpp
index f73104b5ba0..13729a0f240 100644
--- a/backends/aoti/slim/core/test/test_as_strided.cpp
+++ b/backends/aoti/slim/core/test/test_as_strided.cpp
@@ -8,9 +8,9 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/core/SlimTensorView-incl.h>
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor_view_incl.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 
 #ifdef CUDA_AVAILABLE
 #include <cuda_runtime.h>
diff --git a/backends/aoti/slim/core/test/test_permute_reshape.cpp b/backends/aoti/slim/core/test/test_permute_reshape.cpp
index 688245d8be7..34c730996e2 100644
--- a/backends/aoti/slim/core/test/test_permute_reshape.cpp
+++ b/backends/aoti/slim/core/test/test_permute_reshape.cpp
@@ -8,9 +8,9 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/core/SlimTensorView-incl.h>
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor_view_incl.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 
 #ifdef CUDA_AVAILABLE
 #include <cuda_runtime.h>
diff --git a/backends/aoti/slim/core/test/test_slimtensor_basic.cpp b/backends/aoti/slim/core/test/test_slimtensor_basic.cpp
index e23178d4de6..0337cca544a 100644
--- a/backends/aoti/slim/core/test/test_slimtensor_basic.cpp
+++ b/backends/aoti/slim/core/test/test_slimtensor_basic.cpp
@@ -8,8 +8,8 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/core/Storage.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/core/storage.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/core/test/test_slimtensor_copy.cpp b/backends/aoti/slim/core/test/test_slimtensor_copy.cpp
index 6d2ed745446..6c48689619d 100644
--- a/backends/aoti/slim/core/test/test_slimtensor_copy.cpp
+++ b/backends/aoti/slim/core/test/test_slimtensor_copy.cpp
@@ -8,9 +8,9 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/core/Storage.h>
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/core/storage.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/core/test/test_slimtensor_dtypes.cpp b/backends/aoti/slim/core/test/test_slimtensor_dtypes.cpp
index 8ecb8d977b7..6883ea5414b 100644
--- a/backends/aoti/slim/core/test/test_slimtensor_dtypes.cpp
+++ b/backends/aoti/slim/core/test/test_slimtensor_dtypes.cpp
@@ -11,7 +11,7 @@
 #include <cstring>
 #include <type_traits>
 
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/core/test/test_storage.cpp b/backends/aoti/slim/core/test/test_storage.cpp
index 5d61019aa2b..d641c6be85a 100644
--- a/backends/aoti/slim/core/test/test_storage.cpp
+++ b/backends/aoti/slim/core/test/test_storage.cpp
@@ -8,7 +8,7 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/Storage.h>
+#include <executorch/backends/aoti/slim/core/storage.h>
 
 #ifdef CUDA_AVAILABLE
 #include <cuda_runtime.h>
diff --git a/backends/aoti/slim/cuda/guard.h b/backends/aoti/slim/cuda/guard.h
new file mode 100644
index 00000000000..f553e0dcc4b
--- /dev/null
+++ b/backends/aoti/slim/cuda/guard.h
@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and 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.
+ */
+
+#pragma once
+
+#ifdef CUDA_AVAILABLE
+
+#include <cuda.h>
+#include <cuda_runtime.h>
+
+#include <executorch/backends/aoti/slim/c10/core/Device.h>
+#include <executorch/backends/aoti/slim/c10/cuda/Exception.h>
+
+namespace executorch::backends::aoti::slim::cuda {
+
+/**
+ * CUDAGuard - RAII class that sets the current CUDA device.
+ *
+ * This class saves the current CUDA device on construction and restores it
+ * on destruction, providing exception-safe device switching.
+ *
+ * Thread Safety: NOT THREAD-SAFE
+ * - Must only be used within a single thread
+ */
+struct CUDAGuard {
+  /// No default constructor - device must be specified.
+  CUDAGuard() = delete;
+
+  /// Sets the current CUDA device to the specified device index.
+  /// @param device_index The CUDA device index to switch to.
+  explicit CUDAGuard(c10::DeviceIndex device_index) {
+    set_index(device_index);
+  }
+
+  /// Sets the current CUDA device to the specified device.
+  /// @param device The CUDA device to switch to. Must be a CUDA device.
+  explicit CUDAGuard(c10::Device device) {
+    ET_CHECK_MSG(device.is_cuda(), "Expected a CUDA device for CUDAGuard");
+    set_index(device.index());
+  }
+
+  // Copy is not allowed
+  CUDAGuard(const CUDAGuard&) = delete;
+  CUDAGuard& operator=(const CUDAGuard&) = delete;
+
+  // Move is not allowed
+  CUDAGuard(CUDAGuard&& other) = delete;
+  CUDAGuard& operator=(CUDAGuard&& other) = delete;
+
+  /// Restores the original CUDA device on destruction.
+  ~CUDAGuard() {
+    if (original_device_index_ != current_device_index_) {
+      ET_CUDA_LOG_WARN(cudaSetDevice(original_device_index_));
+    }
+  }
+
+  /// Sets the CUDA device to the given device index.
+  /// @param device_index The device index to switch to.
+  void set_index(c10::DeviceIndex device_index) {
+    int orig_index = -1;
+    ET_CUDA_CHECK(cudaGetDevice(&orig_index));
+
+    original_device_index_ = orig_index;
+    current_device_index_ = device_index;
+    if (current_device_index_ != original_device_index_) {
+      ET_CUDA_CHECK(cudaSetDevice(current_device_index_));
+    }
+  }
+
+ private:
+  c10::DeviceIndex original_device_index_;
+  c10::DeviceIndex current_device_index_;
+};
+
+} // namespace executorch::backends::aoti::slim::cuda
+
+#endif // CUDA_AVAILABLE
diff --git a/backends/aoti/slim/factory/Empty.h b/backends/aoti/slim/factory/empty.h
similarity index 94%
rename from backends/aoti/slim/factory/Empty.h
rename to backends/aoti/slim/factory/empty.h
index c0ab9d7248d..b48ce223c5b 100644
--- a/backends/aoti/slim/factory/Empty.h
+++ b/backends/aoti/slim/factory/empty.h
@@ -10,9 +10,9 @@
 
 #include <cstdint>
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
-#include <executorch/backends/aoti/slim/util/SizeUtil.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
+#include <executorch/backends/aoti/slim/util/size_util.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/factory/FromBlob.h b/backends/aoti/slim/factory/from_blob.h
similarity index 95%
rename from backends/aoti/slim/factory/FromBlob.h
rename to backends/aoti/slim/factory/from_blob.h
index b0c659419e9..4e9fda25abf 100644
--- a/backends/aoti/slim/factory/FromBlob.h
+++ b/backends/aoti/slim/factory/from_blob.h
@@ -8,9 +8,9 @@
 
 #pragma once
 
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
-#include <executorch/backends/aoti/slim/util/SizeUtil.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
+#include <executorch/backends/aoti/slim/util/size_util.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/slim/factory/targets.bzl b/backends/aoti/slim/factory/targets.bzl
index 5b10967e166..b26549c01a2 100644
--- a/backends/aoti/slim/factory/targets.bzl
+++ b/backends/aoti/slim/factory/targets.bzl
@@ -7,7 +7,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "empty",
         headers = [
-            "Empty.h",
+            "empty.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
@@ -20,7 +20,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "from_blob",
         headers = [
-            "FromBlob.h",
+            "from_blob.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
diff --git a/backends/aoti/slim/factory/test/test_empty.cpp b/backends/aoti/slim/factory/test/test_empty.cpp
index 18e7ead14ef..39a39ac0f76 100644
--- a/backends/aoti/slim/factory/test/test_empty.cpp
+++ b/backends/aoti/slim/factory/test/test_empty.cpp
@@ -8,7 +8,7 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 
 #ifdef CUDA_AVAILABLE
 #include <cuda_runtime.h>
diff --git a/backends/aoti/slim/factory/test/test_from_blob.cpp b/backends/aoti/slim/factory/test/test_from_blob.cpp
index 16d43d545f3..a0c8f3cc495 100644
--- a/backends/aoti/slim/factory/test/test_from_blob.cpp
+++ b/backends/aoti/slim/factory/test/test_from_blob.cpp
@@ -8,9 +8,9 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/core/Storage.h>
-#include <executorch/backends/aoti/slim/factory/Empty.h>
-#include <executorch/backends/aoti/slim/factory/FromBlob.h>
+#include <executorch/backends/aoti/slim/core/storage.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
+#include <executorch/backends/aoti/slim/factory/from_blob.h>
 
 #ifdef CUDA_AVAILABLE
 #include <cuda_runtime.h>
diff --git a/backends/aoti/slim/util/ArrayRefUtil.h b/backends/aoti/slim/util/array_ref_util.h
similarity index 100%
rename from backends/aoti/slim/util/ArrayRefUtil.h
rename to backends/aoti/slim/util/array_ref_util.h
diff --git a/backends/aoti/slim/util/SharedPtr.h b/backends/aoti/slim/util/shared_ptr.h
similarity index 100%
rename from backends/aoti/slim/util/SharedPtr.h
rename to backends/aoti/slim/util/shared_ptr.h
diff --git a/backends/aoti/slim/util/SizeUtil.h b/backends/aoti/slim/util/size_util.h
similarity index 99%
rename from backends/aoti/slim/util/SizeUtil.h
rename to backends/aoti/slim/util/size_util.h
index aaa41329b14..a8fa994273d 100644
--- a/backends/aoti/slim/util/SizeUtil.h
+++ b/backends/aoti/slim/util/size_util.h
@@ -12,7 +12,7 @@
 #include <optional>
 #include <vector>
 
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
 #include <executorch/runtime/platform/assert.h>
 
 namespace executorch::backends::aoti::slim {
diff --git a/backends/aoti/slim/util/targets.bzl b/backends/aoti/slim/util/targets.bzl
index 60a7933a861..65930ce1d84 100644
--- a/backends/aoti/slim/util/targets.bzl
+++ b/backends/aoti/slim/util/targets.bzl
@@ -7,7 +7,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "shared_ptr",
         headers = [
-            "SharedPtr.h",
+            "shared_ptr.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
@@ -19,7 +19,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "array_ref_util",
         headers = [
-            "ArrayRefUtil.h",
+            "array_ref_util.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
@@ -31,7 +31,7 @@ def define_common_targets():
     runtime.cxx_library(
         name = "size_util",
         headers = [
-            "SizeUtil.h",
+            "size_util.h",
         ],
         visibility = ["@EXECUTORCH_CLIENTS"],
         exported_deps = [
diff --git a/backends/aoti/slim/util/test/test_size_util.cpp b/backends/aoti/slim/util/test/test_size_util.cpp
index fd1deed1909..95412a05c92 100644
--- a/backends/aoti/slim/util/test/test_size_util.cpp
+++ b/backends/aoti/slim/util/test/test_size_util.cpp
@@ -8,7 +8,7 @@
 
 #include <gtest/gtest.h>
 
-#include <executorch/backends/aoti/slim/util/SizeUtil.h>
+#include <executorch/backends/aoti/slim/util/size_util.h>
 
 namespace executorch::backends::aoti::slim {
 
diff --git a/backends/aoti/tests/test_common_shims_slim.cpp b/backends/aoti/tests/test_common_shims_slim.cpp
index ca744565955..94319c6f94d 100644
--- a/backends/aoti/tests/test_common_shims_slim.cpp
+++ b/backends/aoti/tests/test_common_shims_slim.cpp
@@ -12,7 +12,7 @@
 #include <executorch/backends/aoti/common_shims_slim.h>
 #include <executorch/backends/aoti/slim/c10/core/Device.h>
 #include <executorch/backends/aoti/slim/c10/core/ScalarType.h>
-#include <executorch/backends/aoti/slim/factory/Empty.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
 #include <executorch/runtime/core/error.h>
 #include <executorch/runtime/platform/platform.h>
 
diff --git a/backends/cuda/runtime/shims/memory_slim.cpp b/backends/cuda/runtime/shims/memory_slim.cpp
index 93fd884958c..58bf43b34b0 100644
--- a/backends/cuda/runtime/shims/memory_slim.cpp
+++ b/backends/cuda/runtime/shims/memory_slim.cpp
@@ -8,9 +8,9 @@
 
 #include <executorch/backends/cuda/runtime/shims/memory_slim.h>
 
-#include <executorch/backends/aoti/slim/factory/Empty.h>
-#include <executorch/backends/aoti/slim/factory/FromBlob.h>
-#include <executorch/backends/aoti/slim/util/ArrayRefUtil.h>
+#include <executorch/backends/aoti/slim/factory/empty.h>
+#include <executorch/backends/aoti/slim/factory/from_blob.h>
+#include <executorch/backends/aoti/slim/util/array_ref_util.h>
 #include <executorch/runtime/platform/assert.h>
 
 namespace executorch::backends::cuda {
diff --git a/backends/cuda/runtime/shims/memory_slim.h b/backends/cuda/runtime/shims/memory_slim.h
index ec8b8db14f8..5a0845f243c 100644
--- a/backends/cuda/runtime/shims/memory_slim.h
+++ b/backends/cuda/runtime/shims/memory_slim.h
@@ -11,8 +11,8 @@
 #include <cstdint>
 
 #include <executorch/backends/aoti/export.h>
-#include <executorch/backends/aoti/slim/core/SlimTensor.h>
-#include <executorch/backends/aoti/slim/core/SlimTensorView-incl.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor.h>
+#include <executorch/backends/aoti/slim/core/slim_tensor_view_incl.h>
 #include <executorch/runtime/core/error.h>
 
 namespace executorch::backends::cuda {