The ability to certify the correctness of cyber-physical systems often rely on provisioning resources to account for worst-case behaviors. For example, timing certification necessitates that all software tasks are scheduled to be able to meet their deadlines. However, since the execution times of tasks have wide variances—and they continue to increase with growing software and processor complexity—provisioning resources for the worst case can be very pessimistic and also expensive. In this paper we outline some of our recent efforts to address this problem, and emerging techniques to ensure certification of autonomous cyber-physical systems, while ensuring their efficient implementation.