Stream processing engines have been proposed in the past for handling streaming data coming from data sources. But considering sensor networks, there is a need for an approach that allows stream models to reach also computation-capable constrained embedded devices and to implement storage, exchange and computation on those. We propose a stream model that implements sensor-device data handling. The stream processing abstraction and interface allows small motes to store and process the data locally and to route processed data to consumer streams on-demand. This eliminates the need to code motes operation in lower-level languages, allows easy configuration of operations of different types and saves communication energy. The approach is quite useful in diverse contexts, including wireless sensor networks. We describe the approach and show its advantages experimentally.