With the proliferation of sensor networks and sensor network applications, the overall complexity of such systems is continuously increasing. Sensor networks are now heterogeneous in terms of their hardware characteristics and application requirements even within a single network. In addition, the requirements of currently supported applications are expected to change over time. All of this makes developing, deploying, and optimizing sensor network applications an extremely difficult task. In this paper, we present the architecture of TinyCubus, a flexible and adaptive cross-layer framework for TinyOS-based sensor networks that aims at providing the necessary infrastructure to cope with the complexity of such systems. TinyCubus consists of a cross-layer framework that enables optimizations through cross-layer interactions, a configuration engine that distributes components efficiently by considering the roles of the sensor nodes and provides support to install components dynamically, and a data management framework that selects and adapts both system and data management components. Finally, relevant research challenges associated with the development of each framework are identified and discussed in the paper.