This work introduces a Bayesian smoothing approach for the joint graduation of mortality rates across multiple populations. In particular, dynamical linear models are used to induce smoothness across ages through structured dependence, analogously to how temporal correlation is accommodated in state-space time-indexed models. An essential issue in subnational mortality probabilistic modelling is the lack or sparseness of information for some subpopulations. For many countries, mortality data is severely limited, and approaches based on a single population model can result in high uncertainty in the adjusted mortality tables. Here, we recognize the interdependence within a group of mortality data and pursue the pooling of information across several curves that ideally share common characteristics, such as the influence of epidemics or major economic shifts. Our proposal considers multivariate Bayesian dynamical models with common parameters, allowing for borrowing of information across mortality tables and enabling tests of convergence across populations. We also employ discount factors, typical in DLMs, to regulate smoothness, with varying discounting across ages, ensuring less smoothness at younger ages and greater stability at adult ages. This setup implies a trade-off between stability and adaptability. The discount parameter controls the responsiveness of the fit at older ages to new data. The estimation is fully Bayesian, accommodating all uncertainties in modelling and prediction. To illustrate the effectiveness of our model, we analyse male and female mortality data from England and Wales between 2010 and 2012, obtained from the Office for National Statistics. In scenarios with simulated missing data, our approach showed strong performance and flexibility in pooling information from related populations with more complete data.