Assume we have a number of transducers with limited processing capabilities (time and/or space). Are there theoretical results (proofs) on how these transducers can be combined by some sort of communication into one joint transducer to reach a maximum of processing capabilities? Does this work for universal transducers or only for certain problems?

Example: Let's say I have 3 signal-processing computers and need to combine them to steer a self-driving car. Are there theoretical results on how or how not to do that?

Background: I recently came across Stafford Beer's "Viable Systems Model". In my interpretation, it claims to be such an optimal architecture for combination. But I have only found anecdotal evidence or reference to common sense for most parts of that claim, no formal proof.