The programmable network approach allows processing of passing user data in a network, which is highly suitable especially for multimedia streams processing. However, programming of complex stream processing applications for programmable nodes is not effortless since they usually do not provide sufficient flexibility (both programming flexibility and execution environment flexibility). In this paper we present the programmable network node architecture named DiProNN that is able to accept and run user-supplied programs and/or virtual machines and process them over passing data. All the DiProNN programs are described using our modular programming model based on the workflow principles that takes advantages of DiProNN virtualization and makes programming of complex streaming applications easier. As a possible application we show a sketch implementation of simple MCU (Multipoint Control Unit) used for large videoconferences that profits from DiProNN properties.