Non-Intrusive Load Monitoring (NILM) technique or energy disaggregation is a technique used to detect the appliance’s states and estimate their individual energy consumption, given the aggregated data through the main smart meter. Indeed, energy efficiency is the main goal of the NILM techniques, which can be achieved by providing energy disaggregation feedback to the consumers. Unlike single models where training must be performed for each appliance, this work proposes multi-target disaggregation which is more appropriate due to the drastic reduction of resources when training is performed for all target appliances simultaneously. For this purpose, new hybrid models are proposed by combining well-known deep learning models: Convolutional Neural Network (CNN), Denoising Autoencoder (DAE), Recurrent Neural Network (RNN), and Long Short-Term Memory network (LSTM). An implementation and detailed comparative study is then suggested between the proposed hybrid deep learning models and conventional single models in terms of various performance metrics on the UK-Domestic Appliance-Level Electricity (UKDALE) benchmarking database. The experimental results show that the proposed hybrid models provide the best disaggregation performances for multi-target disaggregation compared to single models. Specifically, the CNN-LSTM and the DAE-LSTM are the best hybrid models with the highest overall F1-score of 78.90% and 72.94% respectively.