Currently, organizations strive for faster, usually near-real time, decision making processes including complex event processing and data stream analysis techniques. However, the inclusion of continuous data streams in such process entails new challenges not present in traditional data management systems. First, the data stream management, relying on a sliding window buffering model to smooth arrival irregularities. Second, the data stream processing, relying on linear or sublinear algorithms to provide near real-time analysis. Recently, many efforts have been put to provide adaptation mechanisms to the multiple, continuous, rapid, time-varying nature of data streams into Data Stream Management Systems (DSMS). Nonetheless, these works do not have a holistic view of the architecture they are part of, missing potential optimizations. Even though some works consider Linked Data streams, overcoming Variety issues, they are not exploiting such semantics for optimization. To this end, in this doctoral project we propose to extend the λ-architecture enabling it with semantic aware self-tuning features for optimal real-time processing. Several data stream characteristics must be considered, e.g. arrival rate, data distribution or schema evolution. Combining those with system requirements, e.g. window buffer size and user requirements, e.g. time constraints; will provide a complete knowledge base to enable run-time tuning.