If your functions are transforming chunks of data into other formats/types, you are already doing what data flow graphs are doing. Generalizing can give much more structured concurrency.