An adaptive distributed embedded system is able to automatize some processes at the same time it modifies its behaviour autonomously and dynamically in response to changing operating conditions. To support adaptivity it is necessary that the underlying Distributed Embedded System (DES) is able to dynamically change the assignment of the processing and network resources. In this regard, the DFT4FTT project aims at providing a complete DES that can support applications with real-time, reliability and adaptivity requirements. This paper describes the first steps towards the design of the task allocation scheme used in the DFT4FTT architecture, responsible for dynamically distributing the workload among the nodes of the DES, taking into account the changes in the environment and in the system itself. This allocation scheme not only provides flexibility from a functional point of view, but also from a fault tolerance point of view. Moreover, its modular design makes it possible to tune the desired level of autonomy in the adaptivity, from a simple support for application reconfiguration to a complete automatic reconfiguration assisted with machine learning algorithms.