The purpose of computer vision methods, machine learning, and deep learning in the age of digital pictures is to extract information from them and develop new knowledge. This permits the use of pictures for the early detection and treatment of a broad spectrum of disorders. Deep neural networks are utilised in dermatology to discriminate between melanoma and non-melanoma pictures. We have highlighted two critical elements in melanoma detection research in this work. The first consideration is how even little changes to the parameters in the dataset affect the accuracy of classifiers. In this situation, we looked into Transfer Learning problems. Based on the findings of the first study, we propose that continuous training-test iterations are required to produce robust prediction models. The second argument is the need for a more adaptable system design that can deal with changes in training datasets. In this context, we suggested creating and deploying a hybrid architecture based on Cloud, Fog, and Edge Computing to deliver a Melanoma Detection service based on clinical and dermoscopic pictures. At the same time, this architecture must cope with the volume of data to be studied by shortening the continuous retrain's running time. Experiments on a single computer and several distribution methods have underlined this point, demonstrating how a distributed strategy ensures output attainment in a considerably more adequate period.