Towards Adaptive GPU Resource Management for Embedded Real-Time Systems