Real-time systems are being used in automobiles, avionics, medical systems, nuclear plants, high-speed rail, and smartphones. Current real-time systems are inherently complex and built with heavy over-provisioning of resources to compromise between safety and functionality. Balancing safety and functionality requirements is required to maximize the performance of a system. Safety systems in safety-critical applications have to be completely separated from the control systems. The conservation of complexity is a justification of separation of concerns. With the growing focus on safety-critical systems, the principle of separation increasingly is important for adaptive embedded systems. Adaptive and reconfigurable embedded systems that integrate safety-critical and non-critical components, or that integrate safety and adaptive behaviors, require separation of concerns to control system complexity. Resource dependencies can be minimized by factoring out the reservation and consumption parts into separate programs. Moreover, communication should be separated from control to improve the level of predictability. The approach of splitting the whole program into a minimal set of resource dependencies makes the programs easier to understand and analyze. They can then be joined in a deterministic manner through specified timed interactions such as timed interfaces. Security is an important related area that can benefit through the separation of concerns. To build high-confidence real-time embedded systems, verification and validation are essential to ensure that the worst cases are evaluated and/or tested. Moreover, emerging technologies such as machine learning and edge-computing-based techniques are proven to be effective to increase the performance, safety, and security of real-time embedded systems.