A normal onetime pad uses modular arithmetic to perform rotations on plaintext. An "e" might rotate to a "b" if the key number is greater than 21. This operation is memory-intensive and difficult to perform by hand. So why not encrypt in a way that uses the basic principle of the onetime pad but in a simpler and faster way? Say you have some text that you want to encrypt. You take a chart (or you could program this whole process into a computer) with the ninety-five printable ASCII characters and an additional five characters, numbered from 00 to 99. You convert the text to a stream of numbers using this chart.
Then you create a key the length of the message using a secure random number generator, or just make one up and write it down. It is easier to use the generator because then you only have to remember the seed rather than the entire key. Anyway, you take the key and subtract it from the text. If you get a negative number it is fine. The answer is the ciphertext.
Since there is an infinite number of pairs of numbers whose difference is the ciphertext, it is impossible to decipher it without the key. A bruteforce search of all keys will yield literally every possible string of text. This means that there will be an infinite number of answers that make perfect sense even though they are wrong.
Anyone could easily do this by hand, and it would be EXTREMELY fast on a computer. Here is an example.

"ABCabcDEF" is converted to 34 35 36 66 67 68 37 38 39.

Using some algorithm or just making it up, you get a key: 12 35 53 19 71 22 66 66 42.

Subtract. 343536666768373839
-123553197122666642
--------------------------------
219983469645707197

If you just have the number 219983469645707197 then there is no way to find the message or the key. Also, the key can be longer or shorter than the message, it does not have to be the same length.
Nobody steal my idea; I'm getting it patented and U will be screwed over.