In this paper, we propose an efficient and robust convolutional autoencoder (CAE) model for continuous realtime blood pressure (BP) monitoring. The proposed model was implemented on a resource-constrained edge device. The model was built to capture the hidden patterns among successive segments and alleviate the effects of momentary glitches and outliers. The model was deployed and assessed on the Arduino Nano 33 BLE Sense in a real-time environment by means of Tiny Machine Learning (TinyML). Extensive results revealed that the proposed model improved BP prediction accuracy on both offline and real-time experiments. With 4 features, the model achieved a mean absolute error±standard deviation (MAE±SD) of 2.81±2.84 and 1.51±1.85 mmHg for systolic BP (SBP) and diastolic BP (DBP), respectively, on a dataset of 40 subjects. Whereas microcontroller unit (MCU) based real-time continuous predictions attained 2.25±2.82 for SBP and 5.01±2.10 mmHg for DBP, on 8 volunteers. Compared to the state-of-the-art models implemented on tiny devices, our model showed superior robustness and accuracy. Overall, the study offered some important insights into the significance of compact and impactful feature set and the effectiveness of the proposed model in a real-time setting.