nvidia-opencl-examples VS OpenCL-Headers

The approach of dynamically loading OpenCL.dll and getting function address to call was proposed by nvidia opencl sdk samples in something like 2009. The original link is https://developer.nvidia.com/opencl but now they are not downloading, but some guys on github have that code samples, for example here https://github.com/sschaetz/nvidia-opencl-examples

Khronos has today released the OpenCL 3.0.14 maintenance update that introduces a new cl_khr_command_buffer_multi_device provisional extension that enables execution of a heterogeneous command-buffers across multiple devices. This release also includes significant improvements to the OpenCL C++ Bindings, a new code generation framework for the OpenCL extension headers, and the usual clarifications and bug fixes. The new specifications can be downloaded from the OpenCL Registry.

Basically, it is harder to start without vendor SDK, but it is totally possible. If you have a device with claimed opencl support the only thing you have to install is their vendor-provided driver. Usually (95%) that will provide you an OpenCL.dll / libopencl.so which you can load dynamically, get address of needed API function and use that in your code. To develop from C/C++ environment it is enough to have dynamic library in your system and https://github.com/KhronosGroup/OpenCL-Headers from here.