Organisation and control of heterogeneous cyber-physical systems are complex tasks, that require robust decision making heuristics based on specific knowledge about the physical environment as well as regarding the capabilities of the system's actuators. Hence, fundamentally different concepts exist. Designs based on a centralised hierarchy on the one hand and completely edge-computing based autonomous systems on the other are well investigated and documented but offer specific disadvantages concerning adaptiveness, control and scalability. This article presents a hybrid approach by suggesting an application-aware optimised compromise. Existing approaches as well as requirements and resulting challenges are outlined. Based on that, an extended middleware is introduced, that does not only act as an abstract communication platform but also organises the cyber-physical system by providing a centralised high-level controller component that issues recommendations that are created based on a global view on the domain-specific conditions. Its architecture makes this component completely independent of the lower system levels. Hence, it can flexibly be integrated into different topologies and ported to various use cases. Although the presented approach does not restrict any task allocation strategy, it is shown that decentralised task distribution provides advantages and can easily be implemented based on our recommendations. A simulation-based evaluation demonstrates that this abstract control strategy is able to support performant and effective behaviour of the cyber-physical system.