The success of deep learning (DL) in offline brain-computer interfaces (BCIs) has not yet translated into efficient online applications. This is due to two limiting factors: the need for large amounts of training data in DL and the fact that current DL solutions are primarily developed for offline decoding. To enable real-time decoding, even across subjects without calibration data, we first introduce a novel method real-time adaptive pooling (RAP) to tune existing offline DL models towards online decoding by modifying the pooling layers. To reduce the amount of training data required, our approach leverages different levels of domain adaptation. We show how different settings enable different adaptation solutions. Our results demonstrate that our approach is both powerful and can be calibration-free, providing a robust and practical solution for real-time BCI applications. These findings pave the way for the development of co-adaptive and highly efficient DL-based BCI systems.