gpuArray in Simulink model

I have a simulink model, working with hude graphical arrays. I tried to speed it up, but got a problem - basically Simulink and Stateflow blocks cant pass data of gpuArray type through input/output ports => i must use gather on every intergration step. Gather costs too much - i got only slow down of my model. Is there a way to pass gpuArrays between Simulink blocks?