Membrane proteins are those biomolecules that are attached to or incorporated into the membranes of cells and their organelles. Depending on their functions, they are located in various regions of a cell and are essential to several cellular processes. The locale revelation of these biomolecules is critical as it portrays their activities. Most protein subcellular localization predictors have been trained particularly on globular type and perform poorly on those residing on membranes, specifically through Deep Learning. To overcome this issue, membrane proteins are forecasted in three distinct locations, (a) plasma membrane, (b) internal membrane, and (c) organelle membrane. Features are extracted through Pseudo Amino Acid Composition and some other features from a redundancy curtailed MemLoci dataset. Pseudo Amino Acid Composition is an illustrious approach that excerpts factual protein information through amino acid sequences. Another key feature is that the Pseudo Amino Acid Composition’s impact is unrelated to the Deep Learning Execution of these membrane proteins. This novel study employs four deep learning models, including (a) Artificial Neural Networks (ANN), (b) Recurrent neural networks (RNN), (c) Convolutional neural networks (CNN), and (c) Long Short Term Memory (LSTM). After extensive experimentation, the accuracy of yields is 83.2%, 83.4%, 82.4%., and 80.5% respectively. The outcomes indicate that the simple RNN and ANN models, which are less used in the research, are more suitable compared to the other two models CNN and LSTM which are frequently implemented models in proteomics. The results of the first two models is approximately similar, with a difference of 0.2% among each other, however, they surpass the other two models with better of outcomes in the range of 2 - 3%.