In this contribution, KLT kernels of length 8, 16, and 32 are proposed, which can be applied as separable primary transforms on top of adaptive MTS. These KLTs are commonly applied to every block shape of which width and height ranges from 4 to 32 for only intra-coded CU. And, since they can also be applied to only one direction with DCT-2 coupled in the other, up to three pairs can be candidates as follows: (KLT, KLT), (KLT, DCT-2), and (DCT-2, KLT). These pairs are activated by setting a newly employed flag and also an additional index is followed to indicate one of the pairs. The experimental results for the proposed KLTs are reportedly observed that the BD-rates are reduced by 0.10% with 113%/100% encoding/decoding time changes for AI.