As the fields of machine learning and computer vision are developing, facial recognition systems are becoming increasingly popular and are slowly being widely used in various fields like security, surveillance and medicine. This paper presents the design and development of a facial recognition solution that works in a distributed context, given that the devices used for capturing the images do not have the ability to train models capable of achieving good enough accuracy on large amounts of data. Thus, a method is presented in which the detection of human faces and the characteristics extraction are done locally based on a pre-trained FaceNet model. These characteristics are sent to a strong processing unit where a global model is trained and then transferred back to the clients, where it can be used for recognition. Through experimental evaluation, we show that our solution is efficient and exhibits high accuracy values.