Imagine a distributed mediation application consisting of hundreds of thousands of interconnected nodes, collecting data from millions of pervasive sensors, processing data and delivering it to a myriad of business services. This application takes the form of an acyclic, directed graph. Its shape must continually adapt in response to changes in sensor availability, network layout and business objectives. This involves dynamically adding, configuring, migrating and removing graph nodes. A centralised Observer/Controller, or Autonomic Manager (AM), that controls lifecycle operations for the entire graph would neither scale with the system's size and adaptation frequency, nor survive in unpredictable environments. This paper proposes a decentralised solution for enabling mediation applications to self-grow and to self-adapt their shapes and behaviours. In this approach, applications can autonomously grow into instances of a predefined, abstract architectural model and continually adapt to their execution conditions. A proof-of-concept prototype was developed using a Java-based, Service Oriented Component technology - iPOJO / OSGi. Experimental results from a Home Monitoring data-mediation scenario show the applicability and viability of our approach. We believe that the proposed framework will enable applications to autonomously grow and survive in volatile execution environments, over extended time periods.