I think there are a couple problems with this move:
1.) It doesn't develop. Black is still a couple moves from being able to castle where as white is almost completely developed which means that the Black king runs the risk of getting stuck in the center of the board.
2.) Closely related to the first issue: white is most likely going to play c4 soon which is a normal move in the Caro. With the knight on d5, black is going to loss another tempo which he sorely can't afford.
I suppose you could argue that white is also losing a tempo because he has to move the bishop back to d2 but isn't the case since black had to waste a move in order to get white to waste a move.
Qa5+ is a much better way to kick the bishop. In contrast to Nd5, Qa5 facilitates castling, attacks the pawn on h5, the pawn on a2. If white decides to play c3 with the idea of keeping the bishop on the active square, he allows black to keep a very active queen near the side of the board where white usually castles and he has wasted a tempo since white usually wants to play c4 later which leaves the other option of Bd2 when the queen drops back to c7. Note that the double Queen move doesn't lose a tempo. The position is exactly as it would be if white played Bd2 in the first place and black played Qc7.
Why is KND5 not as good a move as the others.