Obstructive Sleep Apnea is a respiratory disorder that can be the origin of fatal heart and neurological health concerns if left untreated. Despite the availability of diagnosis methods, it is still undiagnosed in most cases due to the tiresome and impractical process of Polysomnography, the current medical standard test. In this study, the authors have worked towards finding a viable approach for easy and early diagnosis of sleep apnea using Electrocardiogram signals. The first model of this work was adapted from the Alexnet architecture with modifications done according to the input digitized signals. A Long–Short term memory layer was added to take care of temporal dependency in the dataset. It has shown an accuracy of 90.87%, specificity of 83.43%, and sensitivity of 95.48%. The hybrid architecture has 1.7 million parameters, much less than the Traditional Alexnet architecture. The second model, ApneaNet, has been introduced, which shows remarkable performance with an accuracy of 90.13%, specificity of 82.06%, and sensitivity of 95.14%, using only 0.9 million parameters which reduce the computational power significantly. The Proposed models have been implemented on a dataset split into 35 recordings for training and testing, showing a trailblazing accuracy of 95.69% and 96.37%, respectively. The authors proposed two deep learning models to detect sleep apnea events using Electrocardiography signals which have demonstrated competitive results compared to the state-of-the-art models at a low computational cost. We believe these methods have the potential to be successfully and efficiently used for the real-time detection of Sleep Apnea.