Being largely utilized as alternative sources of energy, photovoltaic (PV) panels are being deployed in many locations. However, those panels are extensively being prone to faults. A careful detection and diagnosis of those faults becomes a must for solar energy engineers, practitioners and decision-makers in order to extract the maximum of energy from them. The integration of artificial intelligence (AI) in the domain of solar energy management has become a trend in the few previous years because of the high potential provided by machine learning (ML) and deep learning (DL) algorithms in handling issues related to PV systems such as control and fault detection and isolation (FDI). In this paper, TinyML models for defect classification of photovoltaic modules based on electroluminescence (EL) images were developed. An open-source platform, Edge Impulse, was used to design and deploy the models. Two convolutional neural network models (mobileNetV1 and mobileNetV2) have been used and compared. A dataset of EL with three anomalies was used to develop the models. The validation test shows an accuracy of 88.44 and 76.38% for respectively MobileNet-V1 and MobileNet-V2. The quantized (Int8) model has been integrated into a microcontroller (Arduino Portenta H7) for real-time application. The obtained results can be considered as promising and can be extended to other case studies with more defects and larger datasets.