The World Health Organization (WHO) claims that every year 17.9 million deaths are attributed to cardiovascular diseases (CVDs), making them the primary cause of death rates. Electrocardiogram (ECG) is considered the fundamental criterion for assessing heart conditions and also capable for arrhythmia detection. With the recent advancements, artificial intelligence (AI) has become a transformative technology in the area of healthcare with accuracy comparable to human experts. However, deep-learning (DL) models have achieved several notable advancements in medical diagnosis. In this work, we considered 1DCNN with an integration of Bi-LSTM for the classification of 12-lead ECGs of 45,152 patients recorded com- prising of 45152 samples to detect 5 types of classification. The proposed model has achieved an accuracy of 98.9% which is significantly higher. However, these models are computationally intensive and demand a large amount of memory for deployment. Therefore, a lightweight pruned model has been developed for on-device implementation for real-time ECG monitoring with minimum inference time. The lightweight model achieved an accuracy of 99.4% which is higher than the original DL model.