Early recognition and classification of retinal diseases play consequential roles in preventing vision impairment and timely medical intervention. This paper presents a deep learning (DL) method for multi-disease retinal image classification on the RFMiD, which is deployed on a low-resource device, i.e., a Raspberry Pi 4. We employ multiple Convolutional Neural Network (CNN) architectures to classify different retinal diseases. First, we preprocess the images to enhance its feature then secondly, models were trained with fine-tuned hyperparameters with a fivefold validation method. Finally, model performance was evaluated using standard performance metrices. The performed experiments using DL models, we show that proposed classification method beats many baseline approaches. We also investigate how the models generalize by evaluating their performance for other categories of diseases and with cross-validation techniques. The proposed method can potentially improve diagnostic accuracy and efficiency in retinal disease diagnosis assisting ophthalmologists in making better clinical decisions.