Modern Cyber-Physical Systems combine distributed embedded devices with computation in cloud. The inclusion of cloud increases the smartness of the systems by allowing computationallyand data-intensive tasks such as complex data analytics, optimization and decision making, learning and predictions. However, an important implication of interacting with the physical world is the presence of real-time requirements. It puts the cloud in the loop and requires the cloud to participate in the overall real-time guarantees, which poses a difficult problem. In this paper, we address the problem of providing real-time guarantees (in particular statistical guarantees on response time) by combining edge-cloud processing with runtime performance awareness and adaptation.