Port OpenCL Scale to new API

Partially resolves: COMPMID-4190

Change-Id: I680dd80fcbe4e7568511792c60a725b2646fa6ff
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5197
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: TeresaARM <teresa.charlinreyes@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

11 files changed, 377 insertions, 246 deletions

diff --git a/Android.bp b/Android.bp
index 0042dcef4c..3468bd90a1 100644
--- a/Android.bp
+++ b/Android.bp
@@ -142,7 +142,6 @@ cc_library_static {
         "src/core/CL/kernels/CLRemapKernel.cpp",
         "src/core/CL/kernels/CLReorgLayerKernel.cpp",
         "src/core/CL/kernels/CLReverseKernel.cpp",
-        "src/core/CL/kernels/CLScaleKernel.cpp",
         "src/core/CL/kernels/CLSelectKernel.cpp",
         "src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp",
         "src/core/CL/kernels/CLSpaceToDepthLayerKernel.cpp",
@@ -386,6 +385,7 @@ cc_library_static {
         "src/core/gpu/cl/kernels/ClPoolingKernel.cpp",
         "src/core/gpu/cl/kernels/ClQuantizationKernel.cpp",
         "src/core/gpu/cl/kernels/ClReshapeKernel.cpp",
+        "src/core/gpu/cl/kernels/ClScaleKernel.cpp",
         "src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp",
@@ -685,6 +685,7 @@ cc_library_static {
         "src/runtime/gpu/cl/operators/ClPooling.cpp",
         "src/runtime/gpu/cl/operators/ClQuantization.cpp",
         "src/runtime/gpu/cl/operators/ClReshape.cpp",
+        "src/runtime/gpu/cl/operators/ClScale.cpp",
         "src/runtime/gpu/cl/operators/ClSoftmax.cpp",
         "src/runtime/gpu/cl/operators/ClSub.cpp",
         "utils/CommonGraphOptions.cpp",
diff --git a/arm_compute/runtime/CL/functions/CLScale.h b/arm_compute/runtime/CL/functions/CLScale.h
index 1739190972..8a67f74bb6 100644
--- a/arm_compute/runtime/CL/functions/CLScale.h
+++ b/arm_compute/runtime/CL/functions/CLScale.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,28 +26,24 @@
 
 #include "arm_compute/core/KernelDescriptors.h"
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/CL/CLRuntimeContext.h"
 #include "arm_compute/runtime/IFunction.h"
-#include "src/core/CL/kernels/CLFillBorderKernel.h"
-#include "src/core/CL/kernels/CLScaleKernel.h"
 
-#include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
-// Forward declarations
 class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 
-/** Basic function to run @ref CLScaleKernel */
+/** Basic function to run  @ref opencl::ClScale */
 class CLScale : public IFunction
 {
 public:
     /** Default Constructor */
     CLScale();
     /** Default Destructor */
-    ~CLScale() = default;
+    ~CLScale();
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLScale(const CLScale &) = delete;
     /** Default move constructor */
@@ -89,9 +85,9 @@ public:
     // Inherited methods overridden:
     void run() override;
 
-protected:
-    std::unique_ptr<CLFillBorderKernel> _border_handler;
-    std::unique_ptr<CLScaleKernel>      _kernel;
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
-}
+} // namespace arm_compute
 #endif /*ARM_COMPUTE_CLSCALE_H */
diff --git a/docs/00_introduction.dox b/docs/00_introduction.dox
index 7fd3af7c94..1bab6e57d1 100644
--- a/docs/00_introduction.dox
+++ b/docs/00_introduction.dox
@@ -193,7 +193,7 @@ v21.02 Public major release
    - @ref CLPadLayerKernel
    - @ref CLROIAlignLayerKernel
    - @ref CLRangeKernel
-   - @ref CLScaleKernel
+   - CLScaleKernel
    - @ref CLSelectKernel
    - @ref CLBitwiseKernel
    - @ref opencl::kernels::ClFloorKernel
@@ -535,7 +535,7 @@ v20.08 Public major release
    - CLCropKernel
  - Added aligh_corner support for nearest neighbor interpolation in:
    - @ref NEScaleKernel
-   - @ref CLScaleKernel
+   - CLScaleKernel
  - New OpenCL kernels / functions:
    - @ref CLMaxUnpoolingLayerKernel
  - New Neon kernels / functions:
@@ -1060,7 +1060,7 @@ v18.11 Public major release
     - @ref CLDeconvolutionLayer
     - @ref CLL2NormalizeLayer
  - Added QASYMM8 support to the following kernels:
-    - @ref CLScaleKernel
+    - CLScaleKernel
     - NEDepthwiseConvolutionLayer3x3Kernel
     - @ref CLPixelWiseMultiplicationKernel
  - Added FP16 support to the following kernels:
diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index 14733899b3..22fc3e10c4 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -85,7 +85,6 @@
 #include "src/core/CL/kernels/CLRemapKernel.h"
 #include "src/core/CL/kernels/CLReorgLayerKernel.h"
 #include "src/core/CL/kernels/CLReverseKernel.h"
-#include "src/core/CL/kernels/CLScaleKernel.h"
 #include "src/core/CL/kernels/CLSelectKernel.h"
 #include "src/core/CL/kernels/CLSpaceToBatchLayerKernel.h"
 #include "src/core/CL/kernels/CLSpaceToDepthLayerKernel.h"
diff --git a/src/core/CL/kernels/CLScaleKernel.h b/src/core/CL/kernels/CLScaleKernel.h
deleted file mode 100644
index a72e3938d9..0000000000
--- a/src/core/CL/kernels/CLScaleKernel.h
+++ /dev/null
@@ -1,93 +0,0 @@
-/*
- * Copyright (c) 2016-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLSCALEKERNEL_H
-#define ARM_COMPUTE_CLSCALEKERNEL_H
-
-#include "arm_compute/core/KernelDescriptors.h"
-#include "src/core/CL/ICLSimple2DKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** Interface for the scale kernel */
-class CLScaleKernel : public ICLSimple2DKernel
-{
-public:
-    /** Initialise the kernel's inputs, output and interpolation policy
-     *
-     * @param[in]  input  Source tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32
-     * @param[out] output Destination tensor. Data types supported: Same as @p input
-     *                    All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
-     * @param[in]  info   @ref ScaleKernelInfo Kernel descriptor to be used to configure.
-     */
-    void configure(const ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info);
-    /** Initialise the kernel's inputs, output and interpolation policy
-     *
-     * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           Source tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32
-     * @param[out] output          Destination tensor. Data types supported: Same as @p input
-     *                             All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
-     * @param[in]  info            @ref ScaleKernelInfo Kernel descriptor to be used to configure.
-     */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info);
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLScaleKernel
-     *
-     * @param[in] input  Source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32
-     * @param[in] output Destination tensor info. Data types supported: Same as @p input
-     *                   All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
-     * @param[in] info   @ref ScaleKernelInfo Kernel descriptor to be used to validate
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ScaleKernelInfo &info);
-    /** Input tensor accessor.
-     *
-     * @return Pointer to input tensor.
-     */
-    const ICLTensor *input() const;
-    /** Output tensor accessor.
-     *
-     * @return Pointer to output tensor.
-     */
-    const ICLTensor *output() const;
-
-    // Inherited methods overridden:
-    BorderSize border_size() const override;
-    void run(const Window &window, cl::CommandQueue &queue) override;
-
-    // Getter for interpolation policy
-    InterpolationPolicy get_interpolation_policy() const
-    {
-        return _interpolation_policy;
-    }
-
-private:
-    InterpolationPolicy _interpolation_policy = InterpolationPolicy::BILINEAR;
-    DataLayout          _data_layout          = DataLayout::UNKNOWN;
-    bool                _align_corners        = false;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLSCALEKERNEL_H */
diff --git a/src/core/CL/kernels/CLScaleKernel.cpp b/src/core/gpu/cl/kernels/ClScaleKernel.cpp
index c2c78c8f6d..0882f29135 100644
--- a/src/core/CL/kernels/CLScaleKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClScaleKernel.cpp
@@ -21,73 +21,67 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLScaleKernel.h"
+#include "src/core/gpu/cl/kernels/ClScaleKernel.h"
 
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/TensorInfo.h"
 #include "src/core/AccessWindowStatic.h"
 #include "src/core/CL/CLValidate.h"
-#include "src/core/CL/ICLKernel.h"
 #include "src/core/helpers/WindowHelpers.h"
-#include "support/StringSupport.h"
-
 #include "src/core/utils/ScaleUtils.h"
-
-#include <set>
-#include <string>
+#include "support/Cast.h"
 
 namespace arm_compute
 {
+namespace opencl
+{
+namespace kernels
+{
 namespace
 {
-inline std::pair<float, float> calculate_scale_factors(const ITensorInfo &input, const ITensorInfo &output, const ScaleKernelInfo &info)
+inline std::pair<float, float> calculate_scale_factors(const ITensorInfo *src, const ITensorInfo *dst, DataLayout data_layout, bool align_corners)
 {
-    const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? input.data_layout() : info.data_layout;
-    const int        idx_width   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
-    const int        idx_height  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
+    const int idx_width  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
+    const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
 
     // Compute the ratio between source width/height and destination width/height
-    const unsigned int input_width   = input.dimension(idx_width);
-    const unsigned int input_height  = input.dimension(idx_height);
-    const unsigned int output_width  = output.dimension(idx_width);
-    const unsigned int output_height = output.dimension(idx_height);
+    const unsigned int src_width  = src->dimension(idx_width);
+    const unsigned int src_height = src->dimension(idx_height);
+    const unsigned int dst_width  = dst->dimension(idx_width);
+    const unsigned int dst_height = dst->dimension(idx_height);
 
-    float wr = arm_compute::scale_utils::calculate_resize_ratio(input_width, output_width, info.align_corners);
-    float hr = arm_compute::scale_utils::calculate_resize_ratio(input_height, output_height, info.align_corners);
+    float wr = arm_compute::scale_utils::calculate_resize_ratio(src_width, dst_width, align_corners);
+    float hr = arm_compute::scale_utils::calculate_resize_ratio(src_height, dst_height, align_corners);
 
     return std::make_pair(wr, hr);
 }
 
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ScaleKernelInfo &info)
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const ScaleKernelInfo &info)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON(output == input);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON(dst == src);
     ARM_COMPUTE_RETURN_ERROR_ON(info.align_corners && !arm_compute::scale_utils::is_align_corners_allowed_sampling_policy(info.sampling_policy));
 
-    float wr = 0.f;
-    float hr = 0.f;
-    std::tie(wr, hr) = calculate_scale_factors(*input, *output, info);
+    float            wr          = 0.f;
+    float            hr          = 0.f;
+    const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? src->data_layout() : info.data_layout;
+    std::tie(wr, hr) = calculate_scale_factors(src, dst, data_layout, info.align_corners);
 
     ARM_COMPUTE_RETURN_ERROR_ON(info.interpolation_policy == InterpolationPolicy::AREA && (wr > 1.f || hr > 1.f));
 
     return Status{};
 }
 
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const ScaleKernelInfo &info, BorderSize &border)
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info, BorderSize &border)
 {
     Window           win{};
     bool             window_changed{};
     unsigned int     num_elems_processed_per_iteration = 0;
-    const DataLayout data_layout                       = info.data_layout == DataLayout::UNKNOWN ? input->data_layout() : info.data_layout;
+    const DataLayout data_layout                       = info.data_layout == DataLayout::UNKNOWN ? src->data_layout() : info.data_layout;
 
     switch(data_layout)
     {
@@ -100,15 +94,15 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 
             num_elems_processed_per_iteration = 4;
             // Configure kernel window
-            win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
-            AccessWindowStatic input_access(input,
+            win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration));
+            AccessWindowStatic input_access(src,
                                             -border.left, -border.top,
-                                            input->dimension(0) + border.right,
-                                            input->dimension(1) + border.bottom);
-            AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
+                                            src->dimension(0) + border.right,
+                                            src->dimension(1) + border.bottom);
+            AccessWindowHorizontal output_access(dst, 0, num_elems_processed_per_iteration);
 
-            output_access.set_valid_region(win, calculate_valid_region_scale(*(input),
-                                                                             output->tensor_shape(),
+            output_access.set_valid_region(win, calculate_valid_region_scale(*src,
+                                                                             dst->tensor_shape(),
                                                                              info.interpolation_policy,
                                                                              info.sampling_policy,
                                                                              info.border_mode == BorderMode::UNDEFINED));
@@ -119,7 +113,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
         case DataLayout::NHWC:
         {
             // Configure kernel window
-            win = calculate_max_window(*output, Steps());
+            win = calculate_max_window(*dst, Steps());
         }
         break;
         default:
@@ -131,84 +125,64 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 }
 } // namespace
 
-BorderSize CLScaleKernel::border_size() const
+BorderSize ClScaleKernel::border_size() const
 {
     return BorderSize(static_cast<size_t>(_data_layout == DataLayout::NCHW));
 }
 
-Status CLScaleKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ScaleKernelInfo &info)
+Status ClScaleKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const ScaleKernelInfo &info)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, info));
-    const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? input->data_layout() : info.data_layout;
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, info));
+    const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? src->data_layout() : info.data_layout;
     BorderSize       border      = BorderSize(static_cast<size_t>(data_layout == DataLayout::NCHW));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), info, border).first);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), info, border).first);
 
     return Status{};
 }
 
-const ICLTensor *CLScaleKernel::input() const
-{
-    return _input;
-}
-
-const ICLTensor *CLScaleKernel::output() const
-{
-    return _output;
-}
-
-void CLScaleKernel::configure(const ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, output, info);
-}
-
-void CLScaleKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info)
+void ClScaleKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info)
 {
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), info));
-    auto padding_info = get_padding_info({ input, output });
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, info));
+    auto padding_info = get_padding_info({ src, dst });
 
-    _input                = input;
-    _output               = output;
-    _interpolation_policy = info.interpolation_policy;
-    _data_layout          = info.data_layout == DataLayout::UNKNOWN ? input->info()->data_layout() : info.data_layout;
-    _align_corners        = info.align_corners;
+    // Info required for the static tuning
+    _info        = info;
+    _data_type   = src->data_type();
+    _data_layout = _info.data_layout == DataLayout::UNKNOWN ? src->data_layout() : _info.data_layout;
 
     float wr = 0.f;
     float hr = 0.f;
-    std::tie(wr, hr) = calculate_scale_factors(*input->info(), *output->info(), info);
-
-    const bool call_quantized_kernel = is_data_type_quantized_asymmetric(input->info()->data_type()) && _interpolation_policy == InterpolationPolicy::BILINEAR;
-
-    const int  idx_width  = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
-    const int  idx_height = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
-    const bool is_nhwc    = _data_layout == DataLayout::NHWC;
+    std::tie(wr, hr) = calculate_scale_factors(src, dst, _data_layout, _info.align_corners);
+    const bool call_quantized_kernel = is_data_type_quantized_asymmetric(src->data_type()) && _info.interpolation_policy == InterpolationPolicy::BILINEAR;
 
     // Compute actual border size
-    BorderSize border = border_size();
+    BorderSize border  = border_size();
+    const bool is_nhwc = _data_layout == DataLayout::NHWC;
 
-    auto interpolation_policy_to_use = _interpolation_policy;
     // Area interpolation behaves as Nearest Neighbour in case of up-sampling
-    if(_interpolation_policy == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f)
+    auto interpolation_policy_to_use = _info.interpolation_policy;
+    if(_info.interpolation_policy == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f)
     {
         interpolation_policy_to_use = InterpolationPolicy::NEAREST_NEIGHBOR;
     }
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info(), info, border);
+    auto win_config = validate_and_configure_window(src, dst, _info, border);
     ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
     ICLKernel::configure_internal(win_config.second);
 
     // Create kernel
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
-    build_opts.add_option("-DCONSTANT_VALUE=" + string_from_pixel_value(info.constant_border_value, input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));
+    build_opts.add_option("-DCONSTANT_VALUE=" + string_from_pixel_value(info.constant_border_value, src->data_type()));
     build_opts.add_option("-DBORDER_SIZE=" + support::cpp11::to_string(border.right));
     build_opts.add_option_if(info.border_mode == BorderMode::REPLICATE, "-DBORDER_MODE_REPLICATE");
-    build_opts.add_option_if(is_nhwc, "-DDEPTH_OUT=" + support::cpp11::to_string(output->info()->dimension(2)));
-    build_opts.add_option_if_else(info.sampling_policy == SamplingPolicy::CENTER, "-DSAMPLING_POLICY_CENTER", "-DSAMPLING_POLICY_TOP_LEFT");
-    build_opts.add_option_if(_align_corners, "-DALIGN_CORNERS");
+    build_opts.add_option_if(is_nhwc, "-DDEPTH_OUT=" + support::cpp11::to_string(dst->dimension(2)));
+    build_opts.add_option_if_else(_info.sampling_policy == SamplingPolicy::CENTER, "-DSAMPLING_POLICY_CENTER", "-DSAMPLING_POLICY_TOP_LEFT");
+    build_opts.add_option_if(info.align_corners, "-DALIGN_CORNERS");
     if(call_quantized_kernel)
     {
-        const UniformQuantizationInfo qinfo = input->info()->quantization_info().uniform();
+        const UniformQuantizationInfo qinfo = src->quantization_info().uniform();
         build_opts.add_option("-DSCALE=" + support::cpp11::to_string(qinfo.scale));
         build_opts.add_option("-DOFFSET=" + support::cpp11::to_string(qinfo.offset));
     }
@@ -217,41 +191,50 @@ void CLScaleKernel::configure(const CLCompileContext &compile_context, const ICL
     std::string kernel_name = "scale_" + interpolation_name;
     kernel_name += call_quantized_kernel ? "_quantized_" : "_";
     kernel_name += lower_string(string_from_data_layout(_data_layout));
-    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
 
-    unsigned int idx = is_nhwc ? 2 * num_arguments_per_4D_tensor() : 2 * num_arguments_per_2D_tensor(); //Skip the input and output parameters
+    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
+    if(is_nhwc)
+    {
+        ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+    }
 
-    const unsigned int input_width  = input->info()->dimension(idx_width);
-    const unsigned int input_height = input->info()->dimension(idx_height);
+    const int          idx_width  = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
+    const int          idx_height = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
+    unsigned int       idx        = is_nhwc ? 2 * num_arguments_per_4D_tensor() : 2 * num_arguments_per_2D_tensor(); //Skip the input and output parameters
+    const unsigned int src_width  = src->dimension(idx_width);
+    const unsigned int dst_height = src->dimension(idx_height);
 
-    _kernel.setArg<float>(idx++, input_width);
-    _kernel.setArg<float>(idx++, input_height);
+    _kernel.setArg<float>(idx++, src_width);
+    _kernel.setArg<float>(idx++, dst_height);
     _kernel.setArg<float>(idx++, wr);
     _kernel.setArg<float>(idx++, hr);
 
+    // Set to enable static tuning
+    _output_x_dim = dst->dimension(0);
+
     // Set config_id for enabling LWS tuning
     _config_id = "scale_";
-    _config_id += (info.border_mode == BorderMode::REPLICATE ? "Bord_rep" : "");
-    _config_id += (info.sampling_policy == SamplingPolicy::CENTER ? "center" : "topleft");
+    _config_id += (_info.border_mode == BorderMode::REPLICATE ? "Bord_rep" : "");
+    _config_id += (_info.sampling_policy == SamplingPolicy::CENTER ? "center" : "topleft");
     _config_id += (is_nhwc ? "nhwc" : "nchw");
     _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(0));
+    _config_id += support::cpp11::to_string(dst->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(1));
+    _config_id += support::cpp11::to_string(dst->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(2));
+    _config_id += support::cpp11::to_string(dst->dimension(2));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(3));
-    if(is_nhwc)
-    {
-        ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
-    }
+    _config_id += support::cpp11::to_string(dst->dimension(3));
 }
 
-void CLScaleKernel::run(const Window &window, cl::CommandQueue &queue)
+void ClScaleKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
     switch(_data_layout)
     {
         case DataLayout::NCHW:
@@ -261,8 +244,8 @@ void CLScaleKernel::run(const Window &window, cl::CommandQueue &queue)
             do
             {
                 unsigned int idx = 0;
-                add_2D_tensor_argument(idx, _input, slice);
-                add_2D_tensor_argument(idx, _output, slice);
+                add_2D_tensor_argument(idx, src, slice);
+                add_2D_tensor_argument(idx, dst, slice);
                 enqueue(queue, *this, slice, lws_hint());
             }
             while(window.slide_window_slice_2D(slice));
@@ -274,8 +257,8 @@ void CLScaleKernel::run(const Window &window, cl::CommandQueue &queue)
             Window slice     = collapsed.first_slice_window_4D();
 
             unsigned int idx = 0;
-            add_4D_tensor_argument(idx, _input, slice);
-            add_4D_tensor_argument(idx, _output, slice);
+            add_4D_tensor_argument(idx, src, slice);
+            add_4D_tensor_argument(idx, dst, slice);
             enqueue(queue, *this, slice, lws_hint());
             break;
         }
@@ -283,4 +266,6 @@ void CLScaleKernel::run(const Window &window, cl::CommandQueue &queue)
             ARM_COMPUTE_ERROR("Data layout not supported");
     }
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClScaleKernel.h b/src/core/gpu/cl/kernels/ClScaleKernel.h
new file mode 100644
index 0000000000..b6eea0620b
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClScaleKernel.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_SCALE_KERNEL_H
+#define ARM_COMPUTE_CL_SCALE_KERNEL_H
+
+#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for the scale kernel */
+class ClScaleKernel : public IClKernel
+{
+public:
+    /** Default constructor */
+    ClScaleKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClScaleKernel);
+
+    /** Initialise the kernel's inputs, output and interpolation policy
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32
+     * @param[out] dst             Destination tensor info. Data types supported: Same as @p src
+     *                             All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in]  info            @ref ScaleKernelInfo Kernel descriptor to be used to configure.
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClScaleKernel
+     *
+     * @param[in] src  Source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32
+     * @param[in] dst  Destination tensor info. Data types supported: Same as @p src
+     *                 All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in] info @ref ScaleKernelInfo Kernel descriptor to be used to validate
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const ScaleKernelInfo &info);
+
+    // Inherited methods overridden:
+    BorderSize border_size() const override;
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+    // Getter for interpolation policy
+    InterpolationPolicy get_interpolation_policy() const
+    {
+        return _info.interpolation_policy;
+    }
+
+    // Getter for data type
+    DataType get_data_type() const
+    {
+        return _data_type;
+    }
+
+    // Getter for output x dimension
+    unsigned int get_output_x_dim() const
+    {
+        return _output_x_dim;
+    }
+
+private:
+    ScaleKernelInfo _info         = ScaleKernelInfo(InterpolationPolicy::NEAREST_NEIGHBOR, BorderMode::UNDEFINED);
+    DataType        _data_type    = DataType::UNKNOWN;
+    DataLayout      _data_layout  = DataLayout::UNKNOWN;
+    unsigned int    _output_x_dim = 0;
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CLSCALEKERNEL_H */
diff --git a/src/runtime/CL/functions/CLScale.cpp b/src/runtime/CL/functions/CLScale.cpp
index 9862d0a1b3..cbd93c1086 100644
--- a/src/runtime/CL/functions/CLScale.cpp
+++ b/src/runtime/CL/functions/CLScale.cpp
@@ -23,17 +23,26 @@
  */
 #include "arm_compute/runtime/CL/functions/CLScale.h"
 
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/runtime/CL/CLScheduler.h"
+#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClScale.h"
 
 namespace arm_compute
 {
+struct CLScale::Impl
+{
+    const ICLTensor                 *src{ nullptr };
+    ICLTensor                       *dst{ nullptr };
+    std::unique_ptr<opencl::ClScale> op{ nullptr };
+};
+
 CLScale::CLScale()
-    : _border_handler(std::make_unique<CLFillBorderKernel>()), _kernel()
+    : _impl(std::make_unique<Impl>())
 {
 }
+CLScale::~CLScale() = default;
 
 void CLScale::configure(ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info)
 {
@@ -42,33 +51,23 @@ void CLScale::configure(ICLTensor *input, ICLTensor *output, const ScaleKernelIn
 
 void CLScale::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const ScaleKernelInfo &info)
 {
-    auto k = std::make_unique<CLScaleKernel>();
-    k->set_target(CLScheduler::get().target());
-    k->configure(compile_context, input, output, info);
-    _kernel = std::move(k);
+    _impl->src = input;
+    _impl->dst = output;
 
-    // Tune kernels
-    CLScheduler::get().tune_kernel_static(*_kernel);
-
-    const DataLayout data_layout = info.data_layout == DataLayout::UNKNOWN ? input->info()->data_layout() : info.data_layout;
-    if(data_layout == DataLayout::NCHW && !_kernel->border_size().empty())
-    {
-        _border_handler->configure(compile_context, input, _kernel->border_size(), info.border_mode, info.constant_border_value);
-    }
+    _impl->op = std::make_unique<opencl::ClScale>();
+    _impl->op->configure(compile_context, input->info(), output->info(), info);
 }
 
 Status CLScale::validate(const ITensorInfo *input, const ITensorInfo *output, const ScaleKernelInfo &info)
 {
-    return CLScaleKernel::validate(input, output, info);
+    return opencl::ClScale::validate(input, output, info);
 }
 
 void CLScale::run()
 {
-    if(!_kernel->border_size().empty())
-    {
-        CLScheduler::get().enqueue(*_border_handler, false);
-    }
-    CLScheduler::get().enqueue(*_kernel);
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+    _impl->op->run(pack);
 }
-
 } // namespace arm_compute
diff --git a/src/runtime/CL/tuners/BifrostTuner.cpp b/src/runtime/CL/tuners/BifrostTuner.cpp
index 7a06de6d1c..fe95829cca 100644
--- a/src/runtime/CL/tuners/BifrostTuner.cpp
+++ b/src/runtime/CL/tuners/BifrostTuner.cpp
@@ -28,6 +28,7 @@
 #include "support/Cast.h"
 
 #include "src/core/gpu/cl/kernels/ClPoolingKernel.h"
+#include "src/core/gpu/cl/kernels/ClScaleKernel.h"
 
 namespace arm_compute
 {
@@ -234,18 +235,18 @@ void tune_pooling_kernel(opencl::kernels::ClPoolingKernel &k)
     k.set_lws_hint(lws_hint);
 }
 
-void tune_scale_kernel(CLScaleKernel &k)
+void tune_scale_kernel(opencl::kernels::ClScaleKernel &k)
 {
     cl::NDRange               lws_hint      = k.lws_hint();
     const GPUTarget           gpu_target    = k.get_target();
-    const DataType            dt            = k.input()->info()->data_type();
+    const DataType            dt            = k.get_data_type();
     const InterpolationPolicy interpolation = k.get_interpolation_policy();
 
     // Configure the local work size for Bifrost, interpolation (bilinear) and datatype F32.
     // The value are obtained via exhaustive autotuning.
     if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72) && (dt == DataType::F32) && (interpolation == InterpolationPolicy::BILINEAR))
     {
-        auto dim_0 = k.output()->info()->dimension(0);
+        const auto dim_0 = k.get_output_x_dim();
         if(dim_0 == 480)
         {
             lws_hint = cl::NDRange(2, 1);
@@ -285,9 +286,9 @@ void BifrostTuner::tune_kernel_static(ICLKernel &kernel)
     {
         tune_pooling_kernel(*utils::cast::polymorphic_downcast<opencl::kernels::ClPoolingKernel *>(&kernel));
     }
-    else if(dynamic_cast<CLScaleKernel *>(&kernel) != nullptr)
+    else if(dynamic_cast<opencl::kernels::ClScaleKernel *>(&kernel) != nullptr)
     {
-        tune_scale_kernel(*utils::cast::polymorphic_downcast<CLScaleKernel *>(&kernel));
+        tune_scale_kernel(*utils::cast::polymorphic_downcast<opencl::kernels::ClScaleKernel *>(&kernel));
     }
 }
 
diff --git a/src/runtime/gpu/cl/operators/ClScale.cpp b/src/runtime/gpu/cl/operators/ClScale.cpp
new file mode 100644
index 0000000000..4730c8a16e
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClScale.cpp
@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClScale.h"
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+#include "src/core/CL/kernels/CLFillBorderKernel.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClScaleKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClScale::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src);
+    // Configure Scale kernel
+    auto k = std::make_unique<kernels::ClScaleKernel>();
+    k->set_target(CLScheduler::get().target());
+    k->configure(compile_context, src, dst, info);
+    _kernel = std::move(k);
+    if(!_kernel->border_size().empty())
+    {
+        auto b = std::make_unique<CLFillBorderKernel>();
+        b->configure(compile_context, src, _kernel->border_size(), info.border_mode, info.constant_border_value);
+        _border_handler = std::move(b);
+    }
+    // Tune kernel
+    CLScheduler::get().tune_kernel_static(*_kernel);
+}
+
+Status ClScale::validate(const ITensorInfo *src, const ITensorInfo *dst, const ScaleKernelInfo &info)
+{
+    return kernels::ClScaleKernel::validate(src, dst, info);
+}
+
+void ClScale::run(ITensorPack &tensors)
+{
+    ARM_COMPUTE_ERROR_ON_MSG(tensors.empty(), "No inputs provided");
+    if(!_kernel->border_size().empty())
+    {
+        CLScheduler::get().enqueue_op(*_border_handler.get(), tensors, false);
+    }
+    CLScheduler::get().enqueue_op(*_kernel.get(), tensors);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClScale.h b/src/runtime/gpu/cl/operators/ClScale.h
new file mode 100644
index 0000000000..6eccb59be8
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClScale.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_SCALE_H
+#define ARM_COMPUTE_CL_SCALE_H
+
+#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to simulate a scale layer. This function calls the following OpenCL kernels:
+ *
+ * -# @ref CLFillBorderKernel (executed if padding size is different from zero)
+ * -# @ref kernels::ClScaleKernel
+ */
+class ClScale : public IClOperator
+{
+public:
+    /** Constructor */
+    ClScale() = default;
+    /** Initialize the function's source, destination, interpolation type and border_mode.
+     *
+     * @param[in]     compile_context The compile context to be used.
+     * @param[in,out] src             Source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32. (Written to only for @p border_mode != UNDEFINED)
+     * @param[out]    dst             Destination tensor info. Data types supported: Same as @p src
+     *                                All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in]     info            @ref ScaleKernelInfo descriptor to be used to configure
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const ScaleKernelInfo &info);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClScale
+     *
+     * @param[in] src  Source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32.
+     * @param[in] dst  Output tensor info. Data type supported: Same as @p src
+     *                 All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
+     * @param[in] info @ref ScaleKernelInfo descriptor to be used to validate
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const ScaleKernelInfo &info);
+
+    // Inherited method overridden
+    void run(ITensorPack &tensors) override;
+
+protected:
+    std::unique_ptr<ICLKernel> _border_handler{ nullptr };
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CLSCALE_H */