MLKit iOS on GPU/NPU or reduce CPU load?

Hi,

For the last couple of days I'm trying to figure out running MLKit for Pose detection on an iPhone's GPU or NPU for higher efficiency than the CPU. From what I understand there is no GPU/NPU processing possible for now, is there any ETA for that? Or maybe there is a way of reducing the CPU load without reducing quality? (Using the .fast model in practice doesn't change the CPU load much, and there are no other visible things to adjust) The reason behind my question is that on longer recognizing streaks (30-60+ minutes) the phone gets noticeably hot which could probably be avoided with processing done on the GPU/NPU.

Thanks