It may be simpler to code in terms of single fields, instead of lines or "records" of CSV fields. Code sample below uses Java8 streams to build up a result; however, it's trivial to convert it to Java7 using an ArrayList<String> to build up the results.

package test;
import java.util.stream.Stream;

public class CSV {
  public static String[] split(final String text) {
    return (text.indexOf('"') < 0)? text.split(",")
                                  : splitCSV(Stream.<String>builder(), text, 0).build().toArray(String[]::new);
  }


  private static Stream.Builder<String> splitCSV(final Stream.Builder<String> accum, final String text, int start) {
    final int length = text.length();
    if (start >= length) return accum;

    final StringBuilder buf = new StringBuilder(length);
    boolean inquote = false;

    if (text.charAt(start) == '"') {
      inquote = true;
      start++;
    }

    int i = start;

  loop:
    for (; i < length; i++) {
      final char c = text.charAt(i);
      switch (c) {
        case '"':
          if (inquote) {
            if (i + 1 < length && text.charAt(i + 1) == '"') {
              i++;
              break;
            }
            else {
              inquote = false;
              continue loop;
            }
          }
          break;

        case '\\':
          if (i + 1 < length) {
            final char next = text.charAt(i + 1);
            switch (next) {
              case 'b': buf.append('\b'); i++; continue loop;
              case 'f': buf.append('\f'); i++; continue loop;
              case 'r': buf.append('\r'); i++; continue loop;
              case 'n': buf.append('\n'); i++; continue loop;
              case 't': buf.append('\t'); i++; continue loop;
              case '"': buf.append('"');  i++; continue loop;
            }
          }
          break;

        case ',':
          if (!inquote) {
              // value termination
            i++;
            break loop;
          }
          break;
      }

      buf.append(c);
    }

    return splitCSV(accum.add(buf.toString()), text, i);
  }
}

I tested it briefly using the Scala interpreter:

scala> import test.CSV
import test.CSV

scala> CSV.split("N/A,N\\A,ThirdField")
res0: Array[String] = Array(N/A, N\A, ThirdField)

scala> CSV.split(""""A quoted value","One with ""embedded"" quotes",One without,"",a,""")
res1: Array[String] = Array(A quoted value, One with "embedded" quotes, One without, "", a)

scala> res1(3)
res2: String = ""

scala> res2.length
res3: Int = 0

scala> CSV.split("""First,second,"third, with a comma",fourth has a \nnewline,"fifth:\tbeing tricky, has ""some real"" testing craziness!"
     | """
     | )
res4: Array[String] =
Array(First, second, third, with a comma, fourth has a
newline, "fifth:  being tricky, has "some real" testing craziness!
")

scala> CSV.split("""One,second has \"escaped\" quotes,third""")
res5: Array[String] = Array(One, second has "escaped" quotes, third)

Test Notes:

1. Scala has triple-quoted strings that allow embedded single quotes, newlines, and backslashes, etc., without escaping.

2. The Scala interpreter prints empty strings as a set of empty quotes (""), but non-empty strings without surrounding quotes.