Many future earth and space science missions will be composed of multiple spacecraft requiring autonomous capabilities for both opportunistic and coordinated science observations. The Adaptive Network Architecture (ANA) is a software framework composed of multiple, heterogeneous software agents designed for real-time operation of constellations or formations of spacecraft. The ANA is built
upon mature terrestrial standards and best practices for software development, including CORBA Component middleware designed for distributed real-time embedded systems. In this paper we present the further development of the ANA's Science Agent to include a hierarchical computational architecture for reconfigurable on-board science processing. The architecture allows for runtime reconfiguration and/or redeployment of software components across a set of processors based on the available computational resources and changes in operating mode. Application of the science data processing framework to the upcoming Magnetospheric Multi-Scale (MMS) mission is also discussed.