A battery management system (BMS) is a crucial component in numerous small-scale to large-scale industrial applications. An accurately estimated State-Of-Charge (SOC) plays a vital role towards efficient management of a battery system. In this work, a combined machine learning framework, consisting of an optimized long short-term memory (LSTM) neural network and a non-parametric Gaussian process regression (GPR) technique is deployed for the SOC estimation. The composite model is trained using the constant current discharge profiles to obtain more generic and scalable results. The numeric input features (voltage, current and temperature) collected from Panasonic 18650 Lithium-Ion battery cell were used. Extensive training and optimization were performed using GPR followed by a surrogate optimization-based LSTM (GPR-SO-LSTM) via parallel pooling in MATLAB environment. Furthermore, an error correction (EC) algorithm was exploited for increased estimation accuracy. The trained model was tested on carefully curated testing data comprising of different discharge current ranges. Moreover, the accuracy of the model was further challenged by testing it with a battery-powered drill machine. Error metrics such as the average root mean square error (RMSE) of the proposed framework came out to be 0.382% which revealed significant performance improvement in comparison with individual GPR and LSTM models with an RMSE of 2.38% and 7.05% respectively.