The introduction of the 3D-HEVC encoding standard has revolutionized the field of 3D and multi-view video by effectively synthesizing 3D video with adequate depth effects using depth map sequences. However, achieving this involves computationally intensive operations, such as the quad-tree partitioning of Intra Coding Units (CUs). Researchers used multiple approaches including heuristic, machine learning and deep learning to address this complexity. We opted to train a Multi-Deep Convolutional Neural Network (MD-CNN) model specifically for depth maps and integrated it into the 3D-HEVC encoder. This modified encoder uses the independent views as a reference to encode the dependent depth map views. This strategy reduces the complexity of the 3D-HEVC video encoder by 70.12% while slightly reducing video compression efficiency by 0.02% and a small Peak signal-to-noise ratio (PSNR) penalty of -0.80 dB.