In an approach for efficient data processing in a mesh network of computing devices, a processor receives data to be processed in the mesh network. A processor distributes a set of computational units to a set of computing devices in the mesh network. A processor creates a computational plan to construct a computational pipeline of the set of computational units. The computational plan considers cost and performance trade-offs by considering limitation of hardware resources, processing speed, accuracy and application requirements. A processor executes the computational plan by sending data to assigned computational units.