This work was developed in the context of the Novel Enablers for Cloud Slices (NECOS) project. This dissertation aimed to propose an orchestrator of cloud-network resources from federated providers. Given the nature of the NECOS project, the orchestrator should manage portions of the resources inside this federation, which were called slices. Due to the diversity of resources that could make up each slice and aiming to maintain the Service Level Agreement of the platformŠs clients towards their end-users, an automatically customizable orchestration strategy was developed, based on machine learning. Recurrent Neural Networks were used to predict future values of a key performance indicator, previously chosen by the client. Based on this prediction, the orchestrator triggered slice resizing actions, both in the sense of increasing the capacity of the slice to accommodate high loads, as well as reducing this capacity to save resources in case of low demand. The creation of the orchestrator prototype allied with the experiments carried out showed that it is possible, viable, and appropriate to use the proposed strategy.