Dynamic GPU mempool support
Currently we decide whether or not to use mempool with a #define flag, which needs to be set in the HAPI Makefile.
We should change it to such that it can be controlled during runtime, e.g.
hapiMalloc(..., True) to use mempool.
At the first hapiMalloc call with mempool, we can then allocate the mempool (instead of GPU Manager initialization time).