Could you rephrase that slightly? I am not sure if you mean it will work or,
There is no speed up over what is already implemented? or,
The Bitcoin Core already sneaks some code onto the GPU, and is already fully implemented? or,
The GPU would be of no advantage, the most parallel it can become can be ran on a 2+ Core CPU as fast as possible (High step complexity overall)?
It is already parallel, it does not use the GPU. Using the GPU effectively will be difficult because there is not that much parallelism available unless you are verifying multiple blocks at once, and communicating with the GPU has huge latencies. Competing with state of the art cpu code will take state of the art GPU code (my own vanitygen code on the cpu is roughly as fast as the published gpu vanitygen).
I think you should try it, a port of libsecp256k1 to the OpenCL would be pretty interesting and useful. If it were faster we'd probably be willing to do the architectural changes (making it possible to verify multiple blocks at once) to enable using it.