The recently discovered ferroelectricity of van der Waals bilayers offers an unconventional route to improve the performance of devices. Key parameters such as switching field and speed depend on the static and dynamic properties of domain walls (DWs). Here we theoretically explore the properties of textures in stacking-engineered ferroelectrics from first principles. Employing a machine-learning potential model, we present results of large-scale atomistic simulations of stacking DWs and Moir\'e structure of boron nitride bilayers. We predict that the competition between the switching barrier of stable ferroelectric states and the in-plane lattice distortion leads to a DW width of the order of ten nanometers. DWs motion reduces the critical ferroelectric switching field of a monodomain by two orders of magnitude, while high domain-wall velocities allow domain switching on a picosecond-timescale. The superior performance compared to conventional ferroelectrics (or ferromagnets) may enable ultrafast and power-saving non-volatile memories. By twisting the bilayer into a stacking Moir\'e structure, the ferroelectric transforms into a super-paraelectric since DWs move under ultralow electric fields.