It is necessary to inspect the health of the lithium-ion batteries to make them work efficiently and last longer in energy storage devices. With complex changes over time and space, it might prove hard to achieve accurate estimates. This paper introduces a new Deep Kernel Gaussian Process (DKGP) approach that merges offline Gaussian processes with real-time spatiotemporal analysis through a novel deep kernel design. The new approach outperforms previous techniques by integrating two forms of Gaussian processes and a deep kernel design to capture complex features and efficiently analyze spatiotemporal data. The DKGP is unique in that it employs deep neural networks to learn the kernel and adaptive Gaussian processes to quantify uncertainty, enhancing prediction accuracy and confidence. Experiments on a high-fidelity dataset consisting of 29 pairs of 24 V lithium iron phosphate battery systems comprising 232 individual batteries and more than 131 million operation records demonstrate outstanding results with 0.038 RMSE and 0.027 MAE, along with stable predictions in a 95% confidence interval. The DKGP performs excellently in discovering wear patterns in batteries and keeping them operational across varied conditions. Compared to conventional single-modal techniques, the technique performs better in the time-space analysis, and the deep kernel design provides higher reliability and adaptability. The advantages of improved tracking of the health of the batteries include precision, stability, and usability across many industries. Future research will involve enhancing calculations and designing transfer learning mechanisms for various applications.